ohio11_grid_kramer_present.pdf Statspack

1
Rich Niemiec (rich@tusc.com), Rolta TUSC (www.rolta.com www.tusc.com)
(Thanks: Sridhar Avantsa, Mark Komine, Andy Mendelsohn, Debbie Migliore, Maria Colgan,
Kamal Talukder. Steven Tyler, Roger Daltrey, Joe Perry, Aerosmith)
Oracle Disclaimer: The following is intended to outline Oracle's general product direction. It is intended for information purposes only, and may not be incorporated
into any contract. It is not a commitment to deliver any material, code, or functionality, and should not be relied upon in making purchasing decisions. The
development, release, and timing of any features or functionality described for Oracle's products remains at the sole discretion of Oracle.
Oracle Database 11g Best New Features,
Grid Tuning, Block Level Tuning
Ohio 2011

2
Audience Knowledge
• Oracle9i Experience ?
• Oracle9i RAC Experience?
• Oracle10g Experience?
• Oracle Database 11g Experience?
• Goals
– Tuning Tips including block level & AWR / Grid Tuning
– Focus on a few nice features of Oracle 10g & 11g
• Non-Goals
– Learn ALL aspects of Tuning Oracle

What if Kramer was your DBA &
Seinfeld Tuned your Database?
“Jerry, I can‟t find my backup”

4
Overview
• Kramer‟s missing Backup
• George‟s Untuned System
• Elaine‟s Untouched System
• Jerry‟s Perfect Tuning Plan
• Statspack / AWR
– Top Waits
– Load Profile
– Latch Waits
– Top SQL
– Instance Activity
– File I/O
• The Future: EM & ADDM
• 11g New Features
• Helpful Block Level, V$/X$ /(Mutexes)
• Summary

5
Kramer doesn‟t have a Backup
My Junior DBA is getting the backup right now!

6
What Kramer did...
• He was logged into production vs. test
• He deleted some production data
• His backup tape was at Jerry‟s apartment
• He taped a Lady Gaga song over the backup
tape.
• He never actually tested the backup so the
older backup tapes don‟t work either
• He doesn‟t have a DR site

7
Jerry reminds Kramer what he could have
done to prevent all of this...
• The backup should have been in a secure location
• With 10g or 11g encrypt the backup so it will
always be protected
• Could have used Oracle‟s Flashback and get deleted
data back
• Data Guard allows you to fail over to a new sight.
• Test your recovery & DR to ensure it will work

8
Jerry reminds Kramer what he could have
done to prevent all of this...
• Just because a database may need to be recovered,
do not delete the "corrupted" database if possible.
– First, take a backup taken of the "corrupt" database.
– If the restore does not work and you did not backup the
"corrupt" database, you may have nothing to work with.
• When trouble shooting a problem query never let
operations reboot the instance.
• Never startup a standby database in normal mode.
• Never apply catpatch to a standby database. Only
run the catpatch on the primary.

9
How to avoid this mistake...
• Kramer could have ensured that he never made this mistake in the first
place by setting up his prompt to show exactly where he was. For
unix/linux, put in your .profile, or .bash_profile.
If using the bash shell:
export PS1="[$(whoami)@$(hostname)]$ “
For sqlplus, it is the sqlprompt setting:
MIKE6:system:mike61> show sqlp
sqlprompt "MIKE6:system:mike61> "
It is setup in a sqlplus admin file:
[oracle@mikek5 oracle]$ tail -2 $ORACLE_HOME/sqlplus/admin/glogin.sql
set sqlprompt "_user 'at' _connect_identifier> "

10
George Doesn‟t Tune Anything
Instead of proactively tuning are you
depending on other means to save you?

11
George doesn‟t Tune Anything...
• George doesn‟t believe in backups; It slows down
the system.
• He uses the “kill -9” for anything slow
• George doesn‟t patch things especially security
• He uses default passwords for speed and so he
doesn‟t have to change any application code.
• He tries not to do anything that requires actual
work to be done
• He never tells anyone that he‟s going to bring down
the system, he just does a “oops, it crashed” with a
Shutdown Abort when he needs to bring it down.

12
Jerry‟s Advice to George...
• Default passwords should be changed at database creation time.
• Good design beats ad-hoc design.
– Don't work in a black box when tuning.
– Establish priorities and work on what is important to the business.
– Set goals so that everyone knows if success is achieved.
• Setting production databases in noarchivelog mode and then relying
on exports or cold backups.
– No way to recover lost data if crash occurs after the backups.
– Recommend turning archivelog mode ON and use RMAN or hot backups
instead.
– Need to validate, test & regularly review backup & DR plans
• Installing Oracle Enterprise Edition downloaded from
technet.oracle.com and not buying any Oracle Support can be a
problem. You can't really patch this.

13
Jerry‟s Advice to George...
• In George‟s system you might find alert log, trace files, or reports
from end users of errors that have been occurring for a long time but
were not addressed. Then the problem either grew into something
larger or due to changes in the business these issues grow into
something much larger.
• Cleanup is not occurring on destination directories (bdump, cdump,
udump, adump).
• Leaves temp to grow a ridiculously huge size. This just trains the
developers to write un-optimized queries.
• NEVER go without a development environment
• George also never tests anything. How times have we been called
in to fix in a production environment that would have been caught
and corrected BEFORE they were a production problem if only
some basic testing had been done?

14
11g changes … fyi only…
When you specify the ORACLE_BASE environment
variable during install, Oracle also uses this value to set
the DIAGNOSTIC_DEST (required for upgrade from
previous version) parameter which includes all ADR
directories (alert, trace, incident, cdump, hm (health
monitor) etc.).
As of Oracle11gR1, the CORE_DUMP_DEST,
BACKGROUND_DUMP_DEST, and
USER_DUMP_DEST are replaced with
DIAGNOSTIC_DEST. In the following output,
ORACLE_BASE is set to u01/app/oracle and
DIAGNOSTIC_DEST is set to /u01/app/oracle/diag:

15
Query V$DIAG_INFO in 11g
select name, value
from v$diag_info;
NAME VALUE
---------------------- -----------------------------------------------
Diag Enabled TRUE
ADR Base /u01/app/oracle
ADR Home /u01/app/oracle/diag/rdbms/o11gb/O11gb
Diag Trace /u01/app/oracle/diag/rdbms/o11gb/O11gb/trace
Diag Alert /u01/app/oracle/diag/rdbms/o11gb/O11gb/alert
Diag Incident /u01/app/oracle/diag/rdbms/o11gb/O11gb/incident
Diag Cdump /u01/app/oracle/diag/rdbms/o11gb/O11gb/cdump
Health Monitor /u01/app/oracle/diag/rdbms/o11gb/O11gb/hm
Default Trace File /u01/app/oracle/diag/rdbms/o11gb/O11gb/trace/O11gb_ora_17776.trc
Active Problem Count 0
Active Incident Count 0
11 rows selected.

16
Elaine Doesn‟t Work past 5 PM
Are you available when they need you?
When you are available, are you easy to deal with?

17
Elaine doesn‟t Work past 5 PM...
• Elaine doesn‟t understand the concept of the DBA.
• She doesn‟t understand the dedication needed
• If users have a problem after 5 PM, they wait until
tomorrow for a solution.
• Elaine doesn‟t have the time each day to do the
required maintenance tasks needed
• Elaine really wants to be (and should be) an Ad-hoc
query user, but wants the salary of a DBA.
• She advises others to never accept a job as DBA, you can
never escape!!! She advises other DBAs to never give
out home#/pager/cell number to developers...

18
Jerry‟s Advice to Elaine...
• Don‟t set the max_dump_file_size set to the default
(unlimited). Can take over file system if trace generated is
bigger than the destination. Can set this in K, M, or G.
• Don‟t use Oracle 8i/9i settings in a 10g or 11g instance. Take
advantage of new features and get rid of backward compatibility
kernel settings.
• She has RAC, but has no redundancy on the private
interconnect. Having the private interconnect compete with
other public network traffic is a bad thing.
• NEVER apply a patch without testing in development first
• NEVER move code into production without test in
development

19
Jerry‟s Advice to Elaine...
• ALWAYS document your environment, changes to the
environment, and custom code – this makes life easier for
those of us who have to support it
• Her datafiles are set to autoextend unlimited. Then you
find out that the file systems they reside on are limited. An
Oracle Error is triggered when Oracle tries to resize and
can't find the space. Clients do this all the time because
they rely on DBCA.
• Don‟t rely on the UNIX administrator‟s word that the
filesystems underneath are I/O tuned. A good DBA
should actually do the LUN recommendation, i.e. R1+0 for
Oracle, R5 for backups, etc. Check File I/O for issues.

20
Jerry is the Productive DBA
Can you stay up 63 hours? You won‟t need to in the future!

21
Jerry‟s Secret to Tuning;
AWR Report and Grid Control
• Jerry is the Master Tuning Expert
• He Knows The Oracle
• He Leverages what he learned in Statspack
• He Learned what‟s new in AWR Report
• He applies his tuning skills to Grid Control
• He Pro-actively tunes to head off problems
• He Re-actively tunes when needed
• He lets Grid Control Tune for him
• He‟s put his knowledge into Grid Control so that he
can be more productive. He‟s running Exadata!

23
Oracle Firsts – Innovation!
1979 First commercial SQL relational database management system
1983 First 32-bit mode RDBMS
1984 First database with read consistency
1987 First client-server database
1994 First commercial and multilevel secure database evaluations
1995 First 64-bit mode RDBMS
1996 First to break the 30,000 TPC-C barrier
1997 First Web database
1998 First Database - Native Java Support; Breaks 100,000 TPC-C
1998 First Commercial RDBMS ported to Linux
2000 First database with XML
2001 First middle-tier database cache
2001 First RDBMS with Real Application Clusters
2004 First True Grid Database
2005 First FREE Oracle Database (10g Express Edition)
2006 First Oracle Support for LINUX Offering
2007 Oracle 11g Released!
2008 Oracle Exadata Server Announced (Oracle buys BEA)
2009 Oracle buys Sun – Java; MySQL; Solaris; Hardware; OpenOffice
2010 Oracle announces MySQL Cluster 7.1, Exadata V2-8, Exalogic
2011 Storage Expansion Rack, Database Appliance, SPARC SuperCluster T4-4

24
• The Focus has been Acquisitions and gaining Market Share
• Oracle 11g Database extends an already large lead
– Easier to Manage the Database – Better Grid Control
– Self Tuning through a variety of tools (Makes 1 person equal 10)
– Better Security/Encryption & Recoverability via Flashback
– Better Testing Tools (Real Application Testing)
• Andy Mendelsohn is still the database lead
• Releases of Siebel, PeopleSoft, JDE and Oracle12 Apps.
• New Oracle BI Suite & Acquisition of Hyperion
• Acquisition of BEA, SUN
In 2007: Version 11g was Released

25
Oracle gets Sun: Java, MySQL, Solaris,
OpenOffice, Hardware, Storage Tech

27
Tuning -
Leverage ALL of your Knowledge
Do Developers think of this when they think of their
Data

28
Tuning in General
• Both an Art and a Science – You make miracles!
• Exceptions often rule the day…Not a “one size fits all”
• Hardware & Architecture must be right for your
application or it will be difficult to succeed.
• Enterprise Manager (also 3rd party products) are best for
simple tuning and ongoing maintenance.
• V$/X$ are best for drilling deep into problems
• 11g Enterprise Manager radically makes you better!

29
Grid Control: Run the AWR Report
(Many examples follow referencing this)

30
Check Regularly
1. Top 5 wait events
2. Load Profile
3. Instance Efficiency Hit Ratios
4. Wait Events
5. Latch Waits
6. Top SQL
7. Instance Activity
8. File I/O
9. Memory Allocation
10. Undo

32
AWR – Waits / Instance Efficiency

33
Statspack (old/free way) –
Top 5 Wait Events (Top is 3T)
Top 5 Timed Events
~~~~~~~~~~~~~~~~~~
% Total
Event Waits Time (s) Ela Time
--------------------------- ------------ ----------- --------
db file sequential read 399,394,399 2,562,115 52.26
CPU time 960,825 19.60
buffer busy waits 122,302,412 540,757 11.03
PL/SQL lock timer 4,077 243,056 4.96
log file switch 188,701 187,648 3.83
(checkpoint incomplete)

34
Top Wait Events
Things to look for…
Wait Problem Potential Fix
Sequential Read Indicates many index reads – tune the
code (especially joins); Faster I/O; Don‟t
over index or overuse indexes.
Scattered Read Indicates many full table scans–index, tune
the code; cache small tables; Faster I/O
Free Buffer Increase the DB_CACHE_SIZE;
shorten the checkpoint; tune the code to
get less dirty blocks, faster I/O,
use multiple DBWR‟s.
Buffer Busy Segment Header – Add freelists (if inserts)
or freelist groups (esp. RAC). Use ASSM.

35
Statspack - Top Wait Events
Buffer Busy Data Block – Separate „hot‟ data; potentially
use reverse key indexes; fix queries to
reduce the blocks popularity, use
smaller blocks, I/O, Increase initrans
and/or maxtrans (this one‟s debatable)
Reduce records per block.
Buffer Busy Undo Header – Add segments
or increase size of segment area (auto undo)
Buffer Busy Undo block – Commit more (not too
much) Larger segments/area.
Try to fix the SQL.

36
Statspack - Top Wait Events
Enqueue - ST Use LMT‟s or pre-allocate large extents
Enqueue - HW Pre-allocate extents above HW (high
water mark.)
Enqueue – TX Increase initrans and/or maxtrans (TX4)
on (transaction) the table or index. Fix
locking issues if TX6. Bitmap (TX4) &
Duplicates in Index (TX4).
Enqueue - TM Index foreign keys; Check application
(trans. mgmt.) locking of tables. DML Locks.

37
Transactions Moving through
Oracle: ITL & Undo Blocks
Why INITRANS Matter!

38
User 1 – Updates Row# 1&2
User 3 updates Row 3
(There are also In Memory Updates (IMU) in 11g)
• User1 updates a row with an
insert/update/delete – an ITL is
opened and xid tracks it in the
data block.
• The xid ties to the UNDO
header block which ties to the
UNDO data block for undo.
• If user2 wants to query the row,
they create a clone and rollback
the transaction going to the undo
header and undo block.
• If user3 wants to update same
row (they wait). If user 3 wants
to update different row then they
open a second ITL with an xid
that maps to an undo header that
maps to an undo block.
Cache Layer
2
Transaction Layer
XID 01
XID 02
2 Row 3
1 Row 1
1 Row 2
Lock Byte
ITL 1
ITL 2
User1
Request
User3
Request
XID 02 ITL 2
Row 3
2 Row 3

Log
Buffers
(Chain
Latch)
n
1
2
3
Shared
Pool
SGA
Cache Buffers
Buffer Cache
Buffer
Header
Hash
Buckets
Only ONE block
on the Hash
Chain!
Only 1 user - EMP1 buffer header
(So far it‟s clean and only 1 copy)

40
Let‟s watch the EMP1 buffer header
select lrba_seq, state, dbarfil, dbablk, tch, flag, hscn_bas,cr_scn_bas,
decode(bitand(flag,1), 0, 'N', 'Y') dirty, /* Dirty bit */
decode(bitand(flag,16), 0, 'N', 'Y') temp, /* temporary bit */
decode(bitand(flag,1536), 0, 'N', 'Y') ping, /* ping (to shared or null) bit */
decode(bitand(flag,16384), 0, 'N', 'Y') stale, /* stale bit */
decode(bitand(flag,65536), 0, 'N', 'Y') direct, /* direct access bit */
decode(bitand(flag,1048576), 0, 'N', 'Y') new /* new bit */
from x$bh
where dbablk = 56650
order by dbablk;
LRBA_SEQ STATE DBARFIL DBABLK TCH FLAG HSCN_BAS
---------- ---------- ---------- ---------- ---------- ---------- ----------
CR_SCN_BAS D T P S D N
---------- - - - - - -
0 1 1 56650 0 35659776 4294967295
0 N N N N N N

Log
Buffers
(Chain
Latch)
n
1
2
3
Shared
Pool
SGA
Cache Buffers
Buffer Cache
Buffer
Header
Hash
Buckets
Hash Chain is
now SIX long!
Five CR and the
one Current.
Many Users inserting/querying
(Many versions of the block)

42
Insert in 6 other sessions & drive x$bh
up to the max of 6 versions of block
---------- ---------- ---------- ---------- ---------- ---------- ----------
---------- - - - - - -
0 3 1 56650 1 524416 0
4350120 N N N N N N
0 3 1 56650 1 524416 0
4350105 N N N N N N
365 1 1 56650 7 33562633 4350121
0 Y N N N N N
0 3 1 56650 1 524416 0
4350103 N N N N N N
0 3 1 56650 1 524416 0
4350089 N N N N N N
0 3 1 56650 1 524288 0
4350087 N N N N N N

43
Why only 6 versions of a Block?
(more on this later…)
select a.ksppinm, b.ksppstvl, b.ksppstdf, a.ksppdesc
from x$ksppi a, x$ksppcv b
where a.indx = b.indx
and substr(ksppinm,1,1) = '_'
and ksppinm like '%&1%'
order by ksppinm;
KSPPINM
-------------------------------------------------------------------------------
KSPPSTVL
-------------------------------------------------------------------------------
KSPPSTDF
---------
KSPPDESC
-------------------------------------------------------------------------------
_db_block_max_cr_dba
6
TRUE
Maximum Allowed Number of CR buffers per dba

45
What are you Waiting on?
Is this your Ad-Hoc Query User or Network Administrator?

46
Statspack – Top 25
• Tuning the top 25 buffer get and top 25 physical
get queries has yielded system performance gains
of anywhere from 5 percent to 5000 percent.
• The SQL section of the statspack report tells you
which queries to potentially tune first.
• The top 10 of your SQL statements should
usually not be more than 10 percent of your
buffer gets or disk reads.

47
Statspack – Top SQL
(Top 2 are 5T & 3T of reads!!)
Buffer Gets Executions Gets per Exec %Total Time(s) Time
(s) Hash Value
--------------- ------------ -------------- ------ ------- --------
627,226,570 117 5,360,910.9 4.7 9627.09 10367.04
Module: JDBC Thin Client
SELECT * FROM (select d1.tablespace_name, d1.owner, d1.segment_t
ype, d1.segment_name, d1.header_file, d1.extents, d1.bytes, d1.b
locks, d1.max_extents , d1.next_extent from sys.dba_segments d1
where d1.segment_type != 'CACHE' and tablespace_name not in (s
elect distinct tablespace_name from sys.dba_rollback_segs) orde
409,240,446 175,418 2,332.9 3.1 ####### 59430.83
Module: ? @sap10ci (TNS V1-V3)
SELECT "TABNAME" , "VARKEY" , "DATALN" , "VARDATA" FROM "KAPOL"
WHERE "TABNAME" = :A0 AND "VARKEY" LIKE :A1 ORDER BY "TABNAME" ,
"VARKEY"

48
AWR – Top SQL
(Top 1 is 2T – Second one only 250M)

49
Statspack - Latch Waits
Latch Free – Latches are low-level queueing mechanisms
(they‟re accurately referred to as mutually exclusion
mechanisms) used to protect shared memory structures
in the System Global Area (SGA).
• Latches are like locks on memory that are very quickly
obtained and released.
• Latches are used to prevent concurrent access to a shared
memory structure.
• If the latch is not available, a latch free miss is recorded.

50
New in 11g – Mutexes
• Oracle uses mutexes (mutual exclusion) instead of library
cache latches and library cache pin latches to protect objects.
We still have the shared pool latch.
• A mutex requires less memory space and fewer instructions.
• Mutexes take advantage of CPU architecture that has
“compare and swap” instructions.
• With a mutex, if I have the resource and you can‟t get it after
trying a specified number of times (spins), you sleep and try
again a very short time later
• In 10g Oracle used mutexes for pins: you could use the
undocumented parameter: _kks_use_mutex_pin=false (not in
11g).

51
“Cursor: Pin S”: related Bugs and when fixed
in Oracle 10g, 11g & 12 (future version)
NBBug Fixed Description
9499302 11.1.0.7.7, 11.2.0.2, 12.1.0.0 Improve concurrent
mutex request handling
9591812 11.2.0.2.2, 12.1.0.0 Wrong wait events in
11.2 ("cursor: mutex S" instead of "cursor: mutex X")
6904068 11.2.0.2 High CPU usage when
there are "cursor: pin S" waits (Windows – fixed 11.2.0.2)
7441165 10.2.0.5/11.2.0.2 Prevent preemption
while holding a mutex (fix only works on Solaris)
88575526 10.2.0.4, 11.1.0.7 Session spins / OERI
after 'kksfbc child completion' wait – (Windows only)

52
Statspack – Latch/Mutex Waits – fyi
Much better in 11g!!
Latch Free –
• Most latch problems are related to:
– The failure to use bind variables: library cache mutex (latch in 10g)
– Slow redo log disks or contention (log file sync)
– Buffer cache contention issues (cache buffers lru chain)
– Hot blocks in the buffer cache (cache buffers chains).
• There are also latch waits related to bugs; check Support for
bug reports if you suspect this is the case
(oracle.com/support).
• When latch miss ratios are greater than 0.5 percent, you
should investigate the issue.
• In memory updates have changed things for the better!

53
Statspack - Latch Waits - fyi
Latch Activity for DB: ORA9I Instance: ora9i Snaps: 1 -2
Pct Avg Wait Pct
Get Get Slps Time NoWait NoWait
Latch Requests Miss /Miss (s) Requests Miss
------------------------ -------------- ------ ------ ------ ------------ ------
KCL freelist latch 4,924 0.0 0
cache buffer handles 968,992 0.0 0.0 0
cache buffers chains 761,708,539 0.0 0.4 21,519,841 0.0
cache buffers lru chain 8,111,269 0.1 0.8 19,834,466 0.1
library cache 67,602,665 2.2 2.0 213,590 0.8
redo allocation 12,446,986 0.2 0.0 0
redo copy 320 0.0 10,335,430 0.1
user lock 1,973 0.3 1.2 0

54
Cursor Sharing - 8.1.6+
If v$sqlarea looks like this:
select empno from rich778 where empno =451572
Use cursor_sharing=force (sqlarea goes to this):
select empno from rich778 where empno =:SYS_B_0

55
Mutex waits…
• A “pin” is when a session wants to re-execute a statement that‟s in the
library cache.
• Cursor: “Pin S wait on X mutex”; the session needs a mutex in share
mode on a resource and someone has it in exclusive mode.
• The library cache mutex serializes access to objects in the library
cache. Every time a SQL or PL/SQL procedure, package, function, or
trigger is executed, this library cache mutex is used to search the
shared pool for the exact statement so that it can be reused.
• Shared pool latch / library cache mutex issues occur when space is
needed in the library cache (loads with library cache load latch)
• You can also investigate the views V$MUTEX_SLEEP and
V$MUTEX_SLEEP_HISTORY for more information.

56
Instance Activity – AWR/Statspack
Statistic Total per Second per Trans
--------------------------------- ------------------ -------------- ----------
branch node splits 7,162 0.1 0.0
consistent gets 12,931,850,777 152,858.8 3,969.5
current blocks converted for CR 75,709 0.9 0.0
db block changes 343,632,442 4,061.9 105.5
db block gets 390,323,754 4,613.8 119.8
hot buffers moved to head of LRU 197,262,394 2,331.7 60.6
leaf node 90-10 splits 26,429 0.3 0.0
leaf node splits 840,436 9.9 0.3
logons cumulative 21,369 0.3 0.0
physical reads 504,643,275 5,965.1 154.9
physical writes 49,724,268 587.8 15.3
session logical reads 13,322,170,917 157,472.5 4,089.4
sorts (disk) 4,132 0.1 0.0
sorts (memory) 7,938,085 93.8 2.4
sorts (rows) 906,207,041 10,711.7 278.2
table fetch continued row 25,506,365 301.5 7.8
table scans (long tables) 111 0.0 0.0
table scans (short tables) 1,543,085 18.2 0.5

57
Instance Activity
Terminology… - fyi only
Statistic Description
Session Logical Reads All reads cached in memory. Includes both
consistent gets and also the db block gets.
Consistent Gets These are the reads of a block that are in the
cache. They are NOT to be confused with
consistent read (cr) version of a block in the
buffer cache (usually the current version is read).
Db block gets These are block gotten to be changed. MUST
be the CURRENT block and not a CR block.
Db block changes These are the db block gets (above) that
were actually changed.
Physical Reads Blocks not read from the cache. Either from
disk, disk cache or O/S cache; there are also
physical reads direct which bypass cache using
Parallel Query (not in hit ratios).

58
File I/O
Tablespace
------------------------------
Av Av Av Av Buffer Av Buf
Reads Reads/s Rd(ms) Blks/Rd Writes Writes/s Waits Wt(ms)
-------------- ------- ------ ------- ------------ -------- ---------- ------
PSAPSTABI
14,441,749 171 7.9 1.0 521,275 6 1,234,608 6.2
PSAPVBAPD
13,639,443 161 6.2 1.7 10,057 0 2,672,470 4.2
PSAPEDII
11,992,418 142 5.3 1.0 83,757 1 4,115,714 4.4
PSAPEDID
10,617,042 125 8.1 1.0 64,866 1 3,728,009 6.4
PSAPROLL
998,328 12 13.2 1.0 8,321,252 98 285,060 65.7
• Reads should be below 14ms

60
11g New Features &
Grid Control

Testing the Future Version
Version 11.1.0.6.0 of the Database
Version 11.2.0.1.0 of the Database for Release 2 Examples

62
Oracle Database 11g Release 1:
Upgrade Paths
Source Database Target Database
9.2.0.4.0 (or higher) 11.1.x
10.1.0.2.0 (or higher) 11.1.x
10.2.0.1.0 (or higher) 11.1.x
Source Database Upgrade Path for
Target Database
Target Database
7.3.3.0.0 (or lower) 7.3.4.x --> 9.2.0.8 11.1.x
8.0.5.0.0 (or lower) 8.0.6.x --> 9.2.0.8 11.1.x
8.1.7.0.0 (or lower) 8.1.7.4 --> 9.2.0.8 11.1.x
9.0.1.3.0 (or lower) 9.0.1.4 --> 9.2.0.8 11.1.x
Direct Upgrade Path
In-Direct Upgrade Path

63
Database Upgrade Assistant
(DBUA)
• Command Line Option to Auto Extend System
Files
• Express Edition Upgrade to others
• Integration with Oracle Database 11g Pre-
upgrade Tool
• Moving Data Files into ASM, SAN, and Other
File Systems
• Oracle Base and Diagnostic Destination
Configuration

64
Database Upgrade Assistant
(DBUA)
• DBUA checks before the upgrade:
• Invalid user accounts or roles
• Invalid data types or invalid objects
• De-supported character sets
• Adequate resources (rollback segments, tablespaces, and
free disk space)
• Missing SQL scripts needed for the upgrade
• Listener running (if Oracle Enterprise Manager Database
Control upgrade or configuration is requested)
• Oracle Database software linked with Database Vault
option. If Database Vault is enabled, Disable Database
Vault before upgrade.

65
The New Version – Life is Good!
$ sqlplus ***/***
SQL*Plus: Release 11.1.0.6.0 - Production on Tue Oct 30 11:21:04 2007
Copyright (c) 1982, 2007, Oracle. All rights reserved.
Connected to:
Oracle Database 11g Enterprise Edition Release 11.1.0.6.0 - Production
With the Partitioning, OLAP, Data Mining and Real Application Testing options
SQL> startup
ORACLE instance started.
Total System Global Area 422670336 bytes
Fixed Size 1300352 bytes
Variable Size 306186368 bytes
Database Buffers 109051904 bytes
Redo Buffers 6131712 bytes
Database mounted.
Database opened.

67
Database Information - UP!
11gR1
Monitor
Database
(UP)
Users are
Definitely
Using it!
We have
an alert –
we logged
on as SYS

68
11gR2
Monitor
Database
(UP)
Users are
Using it!
Click on
the HA
Console
Restart
Enabled (Restart Database, ASM, Listener after restart of Software/Hardware)

69
High Availability Console - 11gR2
Events
that are an
issue
Flash
Recovery
Usage

MEMORY_TARGET
&
Automatic Memory Management

71
Automatic Memory Management (AMM)
MEMORY_TARGET in 11g
• First there was some Automatic Memory Mgmt - 9i
– SGA_MAX_SIZE introduced in 9i – Dynamic Memory
– No more Buffers – DB_CACHE_SIZE
– Granule sizes introduced - _ksm_granule_size
• Then came SGA_TARGET – 10g
– Oracle Applications recommends setting this for SGA
– Set minimums for key values (Data Cache / Shared Pool)
• Now there is MEMORY_TARGET – 11g
– SGA + PGA all in one setting; Still set minimums

72
SGA & PGA will be
MEMORY_TARGET

73
Automatically sized SGA Components that
Use SGA_TARGET
Component Initialization Parameter
Fixed SGA None
Shared Pool SHARED_POOL_SIZE
Large Pool LARGE_POOL_SIZE
Java Pool JAVA_POOL_SIZE
Buffer Cache DB_CACHE_SIZE
Streams Pool STREAMS_POOL_SIZE

74
Manually Sized SGA Components that
Use SGA_TARGET
Component Initialization Parameter
Log buffer LOG_BUFFER (pfile only in 10g)
Keep Pool DB_KEEP_CACHE_SIZE
Recycle Pool DB_RECYCLE_CACHE_SIZE
Block caches DB_nK_CACHE_SIZE
Program Global Area (now in MEMORY_TARGET):
Aggregate PGA PGA_AGGREGATE_TARGET

75
Moving from SGA_TARGET to:
MEMORY_TARGET
SQL> sho parameter target
NAME TYPE VALUE
------------------------------------ ------------- ------------------------------
memory_max_target big integer 0
memory_target big integer 0
pga_aggregate_target big integer 110M
sga_target big integer 250M

76
MEMORY_TARGET
ALTER SYSTEM SET MEMORY_MAX_TARGET=360M SCOPE=SPFILE;
(shutdown/startup)
ALTER SYSTEM SET MEMORY_TARGET=360M SCOPE=SPFILE;
ALTER SYSTEM SET SGA_TARGET=0; (or set a minimum)
ALTER SYSTEM SET PGA_AGGREGATE_TARGET=0; (or set a minimum)
NAME TYPE VALUE
------------------------------------ ------------- ------------------------------
memory_max_target big integer 360M
memory_target big integer 360M
pga_aggregate_target big integer 0
sga_target big integer 0

77
MEMORY_TARGET (set minimums)
ALTER SYSTEM SET SGA_TARGET=200M;
ALTER SYSTEM SET PGA_AGGREGATE_TARGET=100M;
NAME TYPE VALUE
------------------------------------ ------------- ------------------------------
memory_max_target big integer 360M
memory_target big integer 360M
pga_aggregate_target big integer 100M
sga_target big integer 200M

78
MEMORY_TARGET - EM

80
First, A quick review:
Flush Buffer Cache
• The new 10g feature allows the flush of the buffer cache. It is NOT
intended for production use, but rather for system testing purposes.
• This can help you in your tuning needs or as a band-aid if you have „free
buffer‟ waits (there are better ways to fix this like writing more often or
increasing the DB_CACHE_SIZE)
• Note that any Oracle I/O not done in the SGA counts as a physical
I/O. If your system has O/S caching or disk caching, the actual I/O
that shows up as physical may indeed be a memory read outside of
Oracle.
• To flush the buffer cache perform the following:
SQL> ALTER SYSTEM FLUSH BUFFER_CACHE;

81
Flush Buffer Cache
Example
select count(*) from tab1;
COUNT(*)
-----------------
1147
Execution Plan
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------
0 SELECT STATEMENT Optimizer=CHOOSE (Cost=4 Card=1)
1 0 SORT (AGGREGATE)
2 1 TABLE ACCESS (FULL) OF 'TAB1' (TABLE) (Cost=4 Card=1147)
Statistics
----------------------------------------------------------
0 db block gets
7 consistent gets
6 physical reads

82
Flush Buffer Cache
Example
select count(*) from tab1; (Run it again and the physical reads go away)
COUNT(*)
-----------------
1147
Execution Plan
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Statistics
----------------------------------------------------------
0 db block gets
7 consistent gets
0 physical reads

83
Flush Buffer Cache
Example
ALTER SYSTEM FLUSH BUFFER_CACHE;
System altered.
select count(*) from tab1; (Flush the cache and the physical reads are back)
COUNT(*)
-----------------
1147
Execution Plan
-----------------------------------------------------------------------------------------------------------------------------------
Statistics
----------------------------------------------------------
0 db block gets
7 consistent gets
6 physical reads

84
The Result Cache
• Function Results of queries and query fragments can be
cached in memory for future executions.
– Choose calculations that frequently run
– Choose data that does NOT frequently change
• RESULT_CACHE & RELIES_ON clauses
• Takes its memory from the Shared Pool
– Set with RESULT_CACHE_MAX_SIZE
– RESULT_CACHE_MODE=force (auto/manual)
• DBMS_RESULT_CACHE.FLUSH to clear
• Is NOT passed between RAC/Grid nodes
• Check the docs for other Restrictions & Rules!!

85
Result Cache Performance
Example Query (1M Row Test)
select *
from (select *
from (select t.country_name, t.city_name,
sum(t.salary) a_sum, max(t.salary) a_max
from emps t
group by t.country_name, t.city_name)
order by a_max desc)
where rownum < 2;

86
Result Cache
Example Performance
Step 1 - In Session 1-
Executed query without hint and it returned an elapsed
time of 3.80 seconds (not cached).
Step 2 - In Session 2 –
Executed query without hint and it returned an elapsed
time of 3.20 seconds (not cached).

87
Result Cache
Example Performance
Step 3 - In Session 2
Executed query with the RESULT_CACHE hint and it
returned an elapsed time of 3.18 seconds (cache it).
Step 4 - In Session 1
Executed query without the RESULT_CACHE hint,
but with RESULT_CACHE_MODE=force and it
returned an elapsed time of 0.86 seconds (cached!!).

88
Result Cache Example Query
From the Oracle Docs
• The RELIES_ON Clause specifies tables or views
that the Function Results are dependent on.
-- Package specification
CREATE OR REPLACE PACKAGE HR IS
...
type DeptInfoRec IS RECORD (avgSal NUMBER,
numberEmployees NUMBER);
-- Function declaration
FUNCTION GetDeptInfo (dept_id NUMBER) RETURN DeptInfoRec
RESULT_CACHE;
...
END HR;

89
Result Cache Example Query
From the Oracle Docs
PACKAGE BODY HR IS
...
-- Function definition
FUNCTION GetDeptInfo (dept_id NUMBER) RETURN DeptInfoRec
RESULT_CACHE RELIES_ON (EMP);
IS
result DeptInfoRec;
BEGIN SELECT AVG(sal), count(*) INTO result
FROM EMP
WHERE deptno = dept_id;
RETURN result;
END;
...
END HR;

90
The Result Cache – V$ Views
• V$RESULT_CACHE_STATISTICS – Displays
the amount of memory to help you determine
memory currently allocated to the result cache.
Other V$ views:
• V$RESULT_CACHE_MEMORY
• V$RESULT_CACHE_OBJECTS
• V$RESULT_CACHE_DEPENDENCY

91
The Result Cache – FYI Only
Digging Deeper
KSPPINM KSPPSTVL KSPPDESC
-------------------------------------- -------- --------------------------------
_result_cache_auto_execution_threshold 1 result cache auto execution
threshold
_result_cache_auto_size_threshold 100 result cache auto max size allowed
_result_cache_auto_time_threshold 1000 result cache auto time threshold
_result_cache_block_size 1024 result cache block size
_result_cache_bypass FALSE bypass the result cache
_result_cache_hash_buckets 1024 hash bucket count
_result_cache_invalid 0 post-invalidation usage allowance
_result_cache_max_result 100 maximum result size as percent
of cache size
_result_cache_remote_expiration 0 maximum life time (min) for any
result using a remote object
_result_cache_timeout 60 maximum time (sec) a session

92
Tuning Tools – FYI Only
DBMS_XPLAN
• Use DBMS_XPLAN to query the execution plan
– Automatically queries the last plan in PLAN_TABLE
– uses a TABLE() function with another pipelined function
– Operation text truncation might be a problem
– Will give additional information after plan
• Highlight filter vs join conditions, if plan table is current
• Displays warning message of old version plan table is being used
– In 11g, a procedure for SQL Plan Baselines (we‟ll cover these later).
DBMS_XPLAN.DISPLAY_SQL_PLAN_BASELINE (
sql_handle IN VARCHAR2 := NULL,
plan_name IN VARCHAR2 := NULL,
format IN VARCHAR2 := 'TYPICAL') <„BASIC‟/‟ALL‟>
RETURN dbms_xplan_type_table;

93
Tuning Tools – FYI Only
DBMS_XPLAN
DBMS_XPLAN Example:
Select *
from table (dbms_xplan.display);
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------------
--------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost | Pstart| Pstop |
--------------------------------------------------------------------------------------
| 0 | UPDATE STATEMENT | | 328 | 2296 | 2 | | |
| 1 | UPDATE | JOURNAL_LINE | | | | | |
| 2 | PARTITION RANGE ALL| | | | | 1 | 4 |
| 3 | TABLE ACCESS FULL | JOURNAL_LINE | 328 | 2296 | 2 | 1 | 4 |
--------------------------------------------------------------------------------------
Note: cpu costing is off, 'PLAN_TABLE' is old version
11 rows selected

95
The Virtual Column
• The value of the virtual column is a derived expression.
– Can be derived from columns of the same table or from constants
– Can include SQL or user-defined PL/SQL functions
• Virtual column DATA is NOT PHYSICALLY STORED.
• You CAN NOT explicitly write to a virtual column
• You CAN create a PHYSICAL index (result is function-
based index) or partition on a virtual column <unlike a
computed column in SQL Server or other databases>
• If you UPDATE columns of a virtual column and it has an
index, then it will be computed on the UPDATE vs. on the
SELECT (very important from a tuning standpoint).
• Index Organized and External Tables can NOT have virtual
columns.

96
The Virtual Column
create table emp_rich
(empno number(4),
sal number(7,2),
yearly_sal generated always as (sal*12),
deptno number(2));
Table created.
insert into emp_rich(empno, sal, deptno)
select empno, sal, deptno from scott.emp;
14 rows created.

97
The Virtual Column
select * from emp_rich;
EMPNO SAL YEARLY_SAL DEPTNO
------------ ----------- ------------------- ----------------
7369 800 9600 20
7499 1600 19200 30
7521 1250 15000 30
7566 2975 35700 20
7654 1250 15000 30
7698 2850 34200 30
...

99
The Invisible Index
• Set an index to VISIBLE or INVISIBLE
– ALTER INDEX idx INVISIBLE;
– ALTER INDEX idx VISIBLE;
– CREATE INDEX... INVISIBLE;
• Great to turn off indexes for a while when you think
they‟re not being used, but BEFORE you drop them.
• Can use INDEX (to override invisibility) or
NO_INDEX (to override visibility) hints to override
either setting.
• The index IS MAINTAINED during DML
• Great for testing!

100
The Invisible Index
create index deptno_invisible_idx on dept_rich(deptno) invisible;
Index created.
select count(*) from dept_rich where deptno = 30; (doesn‟t see the index)
COUNT(*)
--------------
512
Execution Plan
----------------------------------------------------------
Plan hash value: 3024595593
--------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 2 | 4 (0)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | 2 | | |
|* 2 | TABLE ACCESS FULL| DEPT_RICH | 512 | 1024 | 4 (0)| 0:00:01 |
--------------------------------------------------------------------------------

101
The Invisible Index (set visible)
alter index dept_rich_inv_idx visible;
Index altered.
select count(*) from dept_rich where deptno = 30;
(it does see the index)
COUNT(*)
--------------
512
Execution Plan
----------------------------------------------------------
---------------------------------------------------------------------------------------
---------------------------------------------------------------------------------------
| 1 | SORT AGGREGATE | | 1 | 2 | | |
|* 2 | INDEX RANGE SCAN| DEPT_RICH_INV_IDX | 512 | 1024 |1 (0)| 00:00:01 |
---------------------------------------------------------------------------------------

102
The Invisible Index (set visible)
select /*+ no_index(dept_rich dept_rich_inv_idx) */ count(*)
from dept_rich
where deptno = 30; (forces not using the index with hint)
COUNT(*)
----------
512
Execution Plan
----------------------------------------------------------
--------------------------------------------------------------------------------
--------------------------------------------------------------------------------
| 1 | SORT AGGREGATE | | 1 | 2 | | |
|* 2 | TABLE ACCESS FULL| DEPT_RICH | 512 | 1024 | 4 (0)| 0:00:01 |
--------------------------------------------------------------------------------

103
The Invisible Index (check it)
alter index dept_rich_inv_idx invisible;
Index altered.
select index_name, visibility
from dba_indexes (or go to USER_INDEXES)
where index_name = 'DEPT_RICH_INV_IDX„;
INDEX_NAME VISIBILITY
-------------------------------- ------------------
DEPT_RICH_INV_IDX INVISIBLE

105
Create & Rebuild Index Online
• You can create/rebuild indexes even when doing DML on the base
table, but it‟s better to do during low DML activity.
• Prior to Oracle 11g, this required an exclusive lock at the beginning
and end of the rebuild. This lock could cause DML delays and
performance spike. This lock is no longer required for this operation.
• Rebuild is faster than a DROP and CREATE
• Basic Syntax:
CREATE INDEX index_name ON table (col1,...) ONLINE;
Index created.
ALTER INDEX index_name REBUILD ONLINE;
Index altered.

106
Rebuild Index or Coalesce (FYI)
Coalesce Example from Oracle Doc.

107
Rebuild Index or Coalesce
Rebuild:
• Quickly move index
to another tablespace
• Requires more disk
space
• Creates new index
tree and shrinks
heights
• Change
storage/tblspc w/o
Coalesce
• Can‟t move to another
tablespace
• Requires much less
space than rebuild
• Coalesces leaf blocks
that are in the same
branch
• Quickly frees index
leaf blocks for use

Nice Developer Tools/Improvements
DDL_LOCK_TIMEOUT
PL/SQL Expressions
Simple Integer
New PL/SQL Packages

109
The DDL Lock Timeout
• DDL Statements (Create/Alter/Drop) require exclusive
locks and thus sometimes fail due to bad timing.
• The parameter DDL_LOCK_TIMEOUT specifies the
amount of time (in seconds) the DDL statement will wait for
the lock before timing out and failing.
• The default value is 0, the max value is 100000 (27.77 hours).
• Example:
alter session set DDL_LOCK_TIMEOUT = 30
Session altered.

110
Allow Sequences in
PL/SQL Expressions
• In Previous Versions needed to retrieve the value
of a sequence (CURRVAL / NEXTVAL) by
invoking a cursor (explicit or implicit).
In 11g:
• No cursor is needed so the code is more
efficient.
• For big jobs – Saves MANY cursors

111
Allow Sequences in
PL/SQL Expressions
OLD Way
DECLARE
V_NEW_VAL NUMBER;
BEGIN
SELECT MY_SEQ.NEXTVAL INTO V_NEW_VAL
FROM DUAL;
END;
NEW Way
DECLARE
V_NEW_VAL NUMBER;
BEGIN
V_NEW_VAL := MY_SEQ.NEXTVAL;
END;

112
Simple Integer Data Type
• Oracle added the new SIMPLE_INTEGER data type to be
more efficient than PLS_INTEGER since the operations are
done directly at the hardware level. There is also a built-in
NOT NULL condition for SIMPLE_INTEGER.
• The performance is larger when the
PLS_CODE_TYPE=„NATIVE‟ vs. INTERPRETED
• We used a PL/SQL Block to loop through 1 million times
incrementing a numeric variable by one. We executed the test
for each of these three times.
Results:
NUMBER: 1.26s
PLS_INTEGER: 0.88s
SIMPLE_INTEGER: 0.65s

113
Additional Enhancements
New PL/SQL Packages
• DBMS_AUTO_TASK_ADMIN
• DBMS_COMPARISON
• DBMS_DG
• DBMS_EDITIONS_UTILITIES
• DBMS_HM (Health Monitor)
• DBMS_HPROF
• DBMS_MGD_ID_UTL
• DBMS_NETWORK_ACL_ADMIN
• DBMS_RESCONFIG
• DBMS_RESULT_CACHE
• DBMS_SQLDIAG (SQL Repair)
• DBMS_WORKLOAD_CAPTURE
• DBMS_WORKLOAD_REPLAY
• DBMS_XA
• DBMS_XDBADMIN
• DBMS_XEVENT
• DBMS_XMLDTD
• DBMS_XMLINDEX
• DBMS_XMLTRANSLATIONS
• SDO_RDF
• SDO_RDF_INFERENCE

114
Enhanced PL/SQL Packages
• DBMS_ADVISOR
• DBMS_APPLY_ADM
• DBMS_AQ
• DBMS_AQADM
• DBMS_CAPTURE_ADM
• DBMS_CDC_PUBLISH
• DBMS_CDC_SUBSCRIBE
• DBMS_CQ_NOTIFICATION
• DBMS_DATA_MINING
• DBMS_DATA_MINING_TR
ANSFORM
• DBMS_DATAPUMP
• DBMS_EXPFIL
• DBMS_FLASHBACK
• DBMS_HS_PASSTHR
OUGH
• DBMS_LOB
• DBMS_LOGSTDBY
• DBMS_MGWADM
• DBMS_MVIEW
• DBMS_PREDICTIVE_
ANALYTICS
• DBMS_RESOURCE_M
ANAGER
• DBMS_RLMGR
• DBMS_RULE_ADM
• DBMS_SCHEDULER
• DBMS_SERVER_ALE
RT
• DBMS_SESSION
• DBMS_SPACE
• DBMS_SQL
• DBMS_SQLTUNE
• DBMS_STATS
• DBMS_STREAMS_
ADM
• DBMS_TRACE
• DBMS_UTILITY
• DBMS_WORKLOA
D_REPOSITORY
• DBMS_XDB
• DBMS_XMLSCHE
MA
• DBMS_XPLAN
• UTL_INADDR
• UTL_RECOMP
• UTL_SMTP
• UTL_TCP

Nice DBA Tool
Oracle Secure Files

116
Oracle SecureFiles
High-Performance Large Objects
• High-performance transactional access to large object data
–RFID, DICOM medical, CAD, images, 3D spacial
–low-latency, high throughput, concurrent access
–space-optimized storage
• Protect your valuable data ... Keep large objects in the database!
–transactions
–transparent encryption
–compression and de-duplication
–database-quality security, reliability, and scalability
• Better security, single view and management of data
• Superset of LOB interfaces – easy migration

117
Oracle Secure Files
Better Performance than LOBs…
Read Performance Write Performance
0 .0 1 0 .1 1 10 10 0
Mb/Sec
0 .0 1 0 .1 1 10 10 0
Mb/Sec
File Size (Mb) File Size (Mb)
Secure Files
Linux Files
Secure Files
Linux Files
Adding Files using New Disk Space – 2x fast than LOBs
Adding Files using Deleted Space – 22x faster than LOBs
PL/SQL Reads – 6x Faster than LOBs
Your mileage will vary....

ADDM Enhancements
(Automatic Database Diagnostic Monitor)

119
ADDM enhancements
• Global ADDM so that Diagnostics are done across
the entire cluster
• Emergency ADDM for use when database is hung
• On any granularity
– Database Cluster
– Database Instance
– Specific Target (such as host, ASM...etc.)
• Over a specified time NOT tied to a pair of
snapshots

120
ADDM Briefly
Specific
Database
Instance
We have 5
ADDM
Findings
Check them
Here

121
ADDM Briefly
Top ADDM
Findings
Click a
Single
Timeframe
Let’s
Check the
Hard Parse
Issue

122
ADDM Briefly
Detailed
Info
&
Findings
Add’l
Info

123
ADDM - Run NOW!
A Big
Problem
Occurs
Run
ADDM
NOW!

124
ADDM - Run NOW!
Are you
Sure?
Running

125
ADDM – Run NOW!
Done.
CPU
Issue

126
ADDM – Run NOW!
Detail
on CPU
Issue?
Suggested
Fixes

127
ADDM – Run NOW!
View
The
Report

128
ADDM for RAC
AWR 1 AWR 2 AWR 3
Inst 1 Inst 2 Inst 3
Self-Diagnostic Engine
Database-Level
ADDM
11g
Instance-Level
ADDM
• Performance expert in a box
– Now RAC specialist too!
• Identifies the most “Globally
Significant” performance issues for
the entire RAC database
• Database-wide and instance-level
analysis
• Database-wide analysis of:
– Global cache interconnect issues
– Lock manager congestion issues
– Global resource contention, e.g. IO
bandwidth, hot blocks
– Globally high-load SQL
– Skew in instance response times
• Allows drill down to instances
• Runs proactively every hour when
taking AWR Snapshots (default)

129
ADDM Considerations:
• CPU Bottlenecks
• Undersized Memory Structures – SGA / PGA
• I/O Capacity Issues
• High Load SQL statements
• High Load PL/SQL
• RAC specific issues – Global hot block/interconnect
• Application issues such as parsing, locks...etc.
• Concurrency (buffer busy) or hot object issues
• Configuration issues – Redo, Archive, Checkpoint.

130
SQL Tuning Advisors &
SQL Plan Management (SPM)
Reminder: COLLABORATE 12
April 22-26, 2012
Mandalay Bay Convention Center
Las Vegas, Nevada

131
SQL Plan Management
• SQL Plan Management is a mechanism that records/evaluates execution
plan of SQL statements (good & bad) over time and builds SQL Plan
baselines (replaces stored outlines) of existing plans known to be
efficient.
• Events that cause the need for SQL Plan baselines:
– New version of Oracle (New optimizer version – Use capture replay to test effect)
– Changes to optimizer statistics or data changes
– Schema, application or metadata changes (use SQL Advisor to get suggestions)
– System settings changes (Use SQL Replay to find what works)
– SQL Profile (statistics – data skews & correlated columns) creation
• Stored outlines are deprecated (discouraged) in Oracle Database 11g.
Oracle highly recommends migrating existing stored outlines to SQL
plan baselines. A SQL Profile contains additional STATISTICS for
this SQL statement for the query optimizer to generate a better
execution plan. An outline/baseline contains HINTS for this SQL
statement for query optimizer to generate a better execution plan.

132
SQL Plan Management
Recommends for SQL or Group of SQL statements:
• SQL Profile stores STATISTICS for a SQL statement for
the query optimizer to generate a better execution plan.
• A Stored Outline/SQL Plan Baseline contains HINTS
for this SQL statement for query optimizer to generate a
better execution plan.
• A SQL Plan Baseline should evolve with changes in the
system to analyze good/bad plans over time.
• View these in DBA_PLAN_BASELINES
• You can also export a SQL Tuning Set and import it to new
system. Capture baselines for Tuning Set with DBMS_SPM
(see later slide on entire syntax). Can also use a pack/unpack
function to pack/unpack all plans in a system for transporting.

133
SQL Plan Management
Create a SQL Tuning Set
Tuning
Issue
Create a
Tuning
Set from
Top 10
SQL

134
Top Activity – 11gR2
(same look)

135
SQL Plan Management
Tuning
Set Name
Queries

136
SQL Plan Management
Viewing a SQL Tuning Set
Tuning
Set Name
Queries
& Stats

137
SQL Plan Management
Run the
Tuning
Advisor
on this
SQL
Tuning
Set (STS)
Run it
NOW

138
SQL Plan Management
Results
Select
One query
And click
View

139
SQL Plan Management
Click on any SQL ID
SQL
Text
Waits
&
Statistics

140
SQL Plan Management
SQL
Profile
Will
Help 99%

141
SQL Plan Management
Compare
Before
&
After

142
SQL Plan Control
SQL Profiles stored in the system
SQL
Profiles
SQL
Plan
Baselines

143
SQL Performance Analyzer (SPA)
COLLABORATE 12
April 22-26, 2012
Las Vegas, Nevada

144
SQL Performance Analyzer
• Measure and report on performance before and
after a change! DBMS_SQLTUNE package.
Great for:
• Database Upgrades
• Application Upgrades
• Hardware Changes
• Database or Schema Changes
• Best for SQL Tuning – Especially Batches

145
Easy to run – SQL Focus (Test SGA settings):
• Capture SQL
• Transport SQL
• Create a Replay Task
• Set up the environment to Test
• Make any changes to Test (such as SGA/Optimizer)
• Compare before and after performance
• Tune the problems!

146
Click
Here

147
Guided Workflow

148
Optimizer Upgrade (10g vs. 11g)

149
11gR2 - Options
Upgrade
Options

150
11gR2 – Exadata Simulation
Test a
Tuning
Set that
I’ve used
in the past

151
Job is
running

152
New!
View
PL/SQL

153
Click on
Job after
complete
View
Report

154
Simple
Job so
no
benefit

155
SQL Advisors
Tuning
Advisors
Repair
Advisor

156
SQL Advisors
11gR2 – (same)
Tuning
Advisors
(Access
Advisor -
next)
Repair
Advisor
(later)

SQL Access Advisor
NEW Partition Advisor

158
SQL Access Advisor &
• The SQL Advisor now combines the functionality
of the SQL Tuning Advisor, SQL Access Advisor
and the new Partition Advisor.
– Recommends Partitioning Needs
– Utilize a previous SQL Tuning Set
– Take SQL straight from what‟s currently in the
CACHE.
– Create a hypothetical workload
– SQL Access Advisor checks Indexes, Partitions or
Materialized Views (schema related issues)

159
Step One
Use a SQL
Tuning Set

160
Look at
Partitions
Quick
Solution

161
Step 3
Schedule
it.
Run it
Now!

162
Final
Review

163
Job
Submitted
Job
Running
Now.

164
Improve
I/O
Improve
Execution
Time

165
Repair the Problem “on the fly”
The Business of IT is serving information...
Not giving users ORA-600 errors…
The SQL Repair Advisor

166
SQL Repair Advisor
• Used to Repair Problem SQL – Oracle Errors
• Reloads and recompiles SQL statements to gather
diagnostics information to fix.
• Uses the diagnostic information to repair the
problem SQL statement (DBMS_SQLDIAG)
• Will fix error going through compilation,
execution and trying different routes (could be
a slower route for now) to come up with a
temporary SQL Patch without error until fixed.

167
SQL Repair Advisor –
Go straight from Alerts
Go to the
Database
Instance
Click Alert
(ORA-600)
message
text to see
details

168
SQL Repair Advisor –
View Problem Details
Click on
View
Problem
Details to
go to the
Support
Bench

169
Support Workbench - Details
Click on
SQL
Repair
Advisor

170
Results from SQL Repair Advisor
Click on
View to
Get the
Detail
finding of
the Advisor Note a SQL
Patch (FIX for
the SQL) has
been generated

171
SQL Repair Advisor
Recommendation / Confirmation
Click on
Implement
To accept
the SQL
Patch

172
Other 11g Tuning
New Features

173
Real Application Testing!
Database workload capture and replay

174
Database workload capture and replay
• Used to capture database workload on one system and replay later
on a different system. Useful to compare two different systems.
• Could rival LoadRunner in the future (may be more precise!)
Brief Steps:
• Capture workload on a database even from 10gR2
• Restore the database on a test system to the SCN when capture begins
• Perform upgrade and make changes to the test system as needed
• Preprocess the captured workload if it is not preprocessed
• Configure the test system for replay (I don‟t do this here)
• Replay workload on the restored database (I don‟t have this in this
presentation, but will show some of the screens to do it)
• Great to test upgrade to 11g (Capture 10gR2 then test against 11g)

175
Post-Change Test System
…
Replay
Driver
…
…
…
App Server
App Server
App Server
…
Process Process Process
… Process Process Process
…
Can use Snapshot Standby as
test system
Replay
Driver
Client
Client
…
Client
Pre-Change (could be 9.2.0.8 or 10g
Capture) Production System
Processed
Captured
Workload
Backup
Capture Workload

176
Database Replay
FYI Only – Download to view in detail
Real App
Testing:
Database
Replay

177
Replay Options...
• Synchronized Replay
– Exact Concurrency, commits & data divergence minimal
• Unsynchronized Replay
– Not the same concurrency or commits
– Data divergence can be large depending on load test performed
• Creates Report
– Data Divergence
– Error Divergence
– Performance Divergence

178
Database Replay – 11gR2 (quick look)
FYI Only – Download to view in detail
Real App
Testing:
Database
Replay

179
Adding Nodes the Easy Way in 11gR2
(FYI look only)

180
DB Instance – Software/Support
11gR2 - Deployment Procedures
Easy
RAC
Add
Nodes
(provisioning)

181
One
Click
Extend
Cluster

182
Step by
Step on
What
Happened

Partitioning: (FYI Only)
- Tables can be split into many pieces (10g).
- Only a subset of the data is queried
- All of the data COULD be queried
- Leads to enhanced performance of large tables
- Re-orgs & backups can be done on a partition level
- 4 quick examples follow (many many rules for each)
- WHAT‟S NEW IN ORACLE 11G

184
The Rules – See Partitioning Guide

185
Range Partitioning (V8)
CREATE TABLE DEPT
(DEPTNO NUMBER(2),
DEPT_NAME VARCHAR2(30))
PARTITION BY RANGE(DEPTNO)
(PARTITION D1 VALUES LESS THAN (10) TABLESPACE DEPT1,
PARTITION D2 VALUES LESS THAN (20) TABLESPACE DEPT2,
PARTITION D3 VALUES LESS THAN (MAXVALUE) TABLESPACE
DEPT3);
INSERT INTO DEPT VALUES (1, ‘DEPT 1’);

186
Range Partitioning (8i)
(Multi-Column)
create table cust_sales (
acct_no number(5),
cust_name char(30),
sale_day integer not null,
sale_mth integer not null,
sale_yr integer not null)
partition by range (sale_yr, sale_mth, sale_day)
(partition cust_sales_q1 values less than (1998, 04, 01) tablespace users1,
partition cust_sales_q2 values less than (1998, 07, 01) tablespace users2,
partition cust_sales_qx values less than (maxvalue, maxvalue, maxvalue)
tablespace users4);

187
Hash Partitioning (8i)
(Multi-Column)
create table cust_sales_hash (
acct_no number(5),
cust_name char(30),
sale_day integer not null,
sale_mth integer not null,
sale_yr integer not null)
partition by hash (acct_no)
partitions 4
store in (users1, users2, users3, users4);

188
Composite Partitioning v (8i)
CREATE TABLE test5 (data_item INTEGER, length_of_item INTEGER,
storage_type VARCHAR(30), owning_dept NUMBER,
storage_date DATE) PARTITION BY RANGE (storage_date) SUBPARTITION BY
HASH(data_item) SUBPARTITIONS 4
STORE IN (data_tbs1, data_tbs2,
data_tbs3, data_tbs4) (PARTITION q1_1999 VALUES LESS
THAN (TO_DATE('01-apr-1999', 'dd-mon-yyyy')), PARTITION q2_1999
VALUES LESS THAN (TO_DATE('01-jul-1999', 'dd-mon-yyyy')),
PARTITION q3_1999
VALUES LESS THAN (TO_DATE('01-oct-1999', 'dd-mon-yyyy'))
(SUBPARTITION q3_1999_s1 TABLESPACE data_tbs1,
SUBPARTITION q3_1999_s2 TABLESPACE data_tbs2),
PARTITION q4_1999
VALUES LESS THAN (TO_DATE('01-jan-2000', 'dd-mon-yyyy'))
SUBPARTITIONS 8
STORE IN (q4_tbs1, q4_tbs2, q4_tbs3, q4_tbs4,
q4_tbs5, q4_tbs6, q4_tbs7, q4_tbs8), PARTITION q1_2000
VALUES LESS THAN (TO_DATE('01-apr-2000', 'dd-mon-yyyy')));

189
List Partitioning
(Allowed since 9i)
create table dept_part
(deptno number(2),
dname varchar2(14),
loc varchar2(13))
partition by list (dname)
(partition d1_east values ('BOSTON', 'NEW YORK'),
partition d2_west values ('SAN FRANCISCO', 'LOS ANGELES'),
partition d3_south values ('ATLANTA', 'DALLAS'),
partition d4_north values ('CHICAGO', 'DETROIT'));
Table created.

190
Interval Partitioning – 11g
• This is a helpful addition to range partitioning where Oracle
automatically creates a partition when the inserted value exceeds all
other partition ranges. 11g also has Ref & Virtual Column
Partitioning (covered here as fyi only).
There are the following restrictions:
• You can only specify one partitioning key column, and it must be of
NUMBER or DATE type.
• Interval partitioning is NOT supported for index-organized tables.
• Interval Partitioning supports composite partitioning:
– Interval-range *** Interval-hash *** Interval-list
• You can NOT create a domain index on an interval-partitioned table.

191
CREATE TABLE DEPT_new
(DEPTNO NUMBER(2),
(PARTITION D1 VALUES LESS THAN (10),
PARTITION D2 VALUES LESS THAN (20),
PARTITION D3 VALUES LESS THAN (30));
Table created.
SQL> insert into dept_new values(40, 'test2');
insert into dept_new values(40, 'test2')
*
ERROR at line 1:
ORA-14400: inserted partition key does not map to any partition

192
select segment_name, partition_name
from dba_segments
where segment_name = 'DEPT_NEW„;
SEGMENT_NAME PARTITION_NAME
------------------------------ ----------------------------
DEPT_NEW D1
DEPT_NEW D2
DEPT_NEW D3

193
CREATE TABLE DEPT_NEW2
(DEPTNO NUMBER(2),
INTERVAL(10)
PARTITION D2 VALUES LESS THAN (20),
PARTITION D3 VALUES LESS THAN (30))
Table created.
SQL> insert into dept_new2 values(40, 'test2');
1 row created.

194
insert into dept_new2 values(40,null);
select segment_name, partition_name
from dba_segments
where segment_name = 'DEPT_NEW2'
SEGMENT_NAME PARTITION_NAME
------------------------------ ------------------------------
DEPT_NEW2 D1
DEPT_NEW2 D2
DEPT_NEW2 D3
DEPT_NEW2 SYS_P41
DEPT_NEW2 SYS_P42
DEPT_NEW2 SYS_P43

195
System Partitioning – 11g
• Great when you will insert a lot of data and want to break it into
smaller pieces, but in the same table.You decide what data goes there.
• RAC: One partition per node. Reduce interconnect traffic,
locking…(using it as Workload Management Enabled).
• You can NOT forget to specify the partition or you will get an:
ORA-14701: Partition-extended name or bind variable must be
used for DMLs on tables partitioned by System method.
There are the following restrictions:
• Can NOT be used with index-organized tables
• Can NOT play a part in Composite Partitioning
• Can NOT Split
• Can NOT be used with CREATE AS SELECT…
• Can use: CREATE TABLE …PARTITION BY SYSTEM PARTITIONS n
(where n is 1 to 1024K-1)

196
System Partitioning (11g)
CREATE TABLE DEPT
(DEPTNO NUMBER(2),
PARTITION BY SYSTEM
(PARTITION D1,
PARTITION D2,
PARTITION D3);
INSERT INTO DEPT PARTITION D1 VALUES (1, ‘DEPT 1’);
Partition clause is optional for UPDATE & DELETES, but more efficient if
you can use it (Careful to ensure you‟re doing what you need).

197
Reference Partitioning – 11g
(FYI Only)
• Allows the partitioning of two tables related to one
another by referential constraints. The partitioning key is
resolved through an existing parent-child relationship,
enforced by enabled and active primary key and foreign
key constraints.
• Tables with a parent-child relationship can be logically
equi-partitioned by inheriting the partitioning key from
the parent table without duplicating the key columns. The
logical dependency will also automatically cascade
partition maintenance operations, thus making application
development easier and less error-prone.

198
(FYI Only)

199

200
Partition Compression
• You can now COMPRESS individual partitions
• Compression as high as 3.5 to 1 is possible
• Compressed Tables now support
– DML Statements
– Add and Drop Column
– Partition level COMPRESS or NOCOMPRESS
• ALTER TABLE... COMPRESS (old compress)
• ALTER TABLE... NOCOMPRESS
• Table compression now supported for OLTP
• New Advanced Compression Option (chargeable):
– CREATE TABLE t1 COMPRESS FOR ALL OPERATIONS
Presentation by Mike Messina on Compressing & Costs
Also - Shyam Varan Nath – Honey I shrunk the Data Warehouse

201
CREATE TABLE DEPT_new3
(DEPTNO NUMBER(2),
COMPRESS FOR OLTP
interval(10)
PARTITION D2 VALUES LESS THAN (20) NOCOMPRESS,
PARTITION D3 VALUES LESS THAN (30));
Table created.
• NOCOMPRESS - The table or partition is not compressed. This is the default action
• COMPRESS - Suitable for data warehouse. Compression enabled during direct-path inserts only.
• COMPRESS FOR DIRECT_LOAD OPERATIONS - Same affect as the simple COMPRESS.
• COMPRESS FOR ALL OPERATIONS - Suitable for OLTP systems. Compression for all
operations, including regular DML statements. Requires COMPATIBLE to be set to 11.1.0 or higher.
• COMPRESS FOR OLTP - Suitable for OLTP systems. Enables compression for OLTP operations,
including regular DML statements. Requires COMPATIBLE to be set to 11.1.0 or higher and in 11.2
replaces the COMPRESS FOR ALL OPERATIONS Syntax, but COMPRESS FOR ALL
OPERATIONS syntax still exists and is still valid.

202
1 row created.
1 row created.
1 row created.
1 row created.
1 row created.

203
select table_name, partition_name, compression
from dba_tab_partitions
where table_name = 'DEPT_NEW3„;
TABLE_NAME PARTITION_NAME COMPRESS
------------------------------ ------------------------------ ----------------
DEPT_NEW3 D1 ENABLED
DEPT_NEW3 D2 DISABLED
DEPT_NEW3 D3 ENABLED
DEPT_NEW3 SYS_P64 ENABLED
6 rows selected.

204
Compression History – Timeline
(FYI Only)
• Index Compression since 8i
• Table Compression since 9i
– No Additional License Requirement
– Only for direct inserts
– Compression Not Maintained with updates and normal inserts
– Had to re-org table to re-compress over time.
• 11g Advanced Compression
– Additional License Requirement
– Compression Maintained with all DML activity
– No re-orgs required after initial compression
• 11gR2 – Hybrid Columnar Compression ( with Exadata)

205
Advanced Compression
(FYI Only)
• The Oracle Advanced Compression option
contains the following features:
– Data Guard Network Compression
– Data Pump Compression
(COMPRESSION=METADATA_ONLY does not
require the Advanced Compression option)
– Multiple RMAN Compression Levels (RMAN
DEFAULT COMPRESS does not require the
Advanced Compression option)
– OLTP Table Compression
– SecureFiles Compression and Deduplication. LZO
compression algorithm added. Faster than ZLIB.

206
Advanced Compression & OLTP
(FYI Only)
• Compression is maintained at a block level.
• Maintained through DML operations.
• Compression ratio depends on “RANDOMness” of the data.
• DML Impact depends on “RANDOMness” of the data.
Range of 10 to 30%.
– More visible for bulk operations compared to single row operations.
• Significant Performance gains in selects, primarily due to
reduced block scans.
– Exacts / specifics depend on your compression ratio.
– Table scans expect upto 50% reduction in block reads.
– PK based access, impact not noticeable.
• Must evaluate on a case by case basis.

207
Hybrid Columnar Compression
(FYI Only)
• Exadata Hybrid Columnar Compression (EHCC)
• Compression of data @ a column level and block level
• Maintained in Compression Units (CU), of 32k chunks.
• Good Random Row Access but NOT good for updates
• Excellent Table Scan performance.
• 5x to 30x Compression Ratios to be expected.
• Potential Challenges
– DML performance
– Data load speeds visibly impacted.
– Single row access speeds visibly impacted.

209
• Ability to online redefine tables that have
materialized view logs:
– Tables with materialized view logs can now be redefined
online.
– Materialized view logs are now one of the dependent
objects that can be copied to the interim table with the
DBMS_REDEFINITION.COPY_TABLE_DEPENDENTS
package procedure.
• DBMS_STATS performance has been improved.

210
Large-Scale Data Warehouses*
Feature Usage
Source: Oracle ST Survey
* Oracle Survey

211
Automatic Diagnostic Repository
(ADR)

212
Automatic Diagnostic Repository (ADR)
• Oracle 11g includes a Fault Diagnosability Infrastructure
to prevent, detect, diagnose, resolve issues related to
bugs, corruption, etc.
• When a critical error occurs it is assigned an incident
number and all diagnostic data tagged with this in ADR.
• ADR is a file based repository outside of the database
• ADR helps detect problems proactively
• ADR helps limit the damage of interruptions
• ADR helps reduce problem diagnostic time
• ADR simplifies Oracle Support / Customer interaction
• The ADR also contains Health Reports, Trace Files,
Dump Files, SQL Test Cases and Data Repair Records

213
ADR Directory Structure for a
Database Instance
ADR Base
diag
rdbms
<database name>
<SID>
alert cdump incident trace (others)
Alert Log: /u01/app/oracle/diag/rdbms/o11gb/O11gb/trace
ORACLE_HOME: /u01/app/oracle/product/11.1.0/db_1

214
ADR – V$ Diagnostic Info
select name, value
from v$diag_info;
NAME VALUE
----------------------------- ------------------------------------------------
Diag Enabled TRUE
ADR Base /u01/app/oracle
ADR Home /u01/app/oracle/diag/rdbms/o11gb/O11gb
Diag Trace /u01/app/oracle/diag/rdbms/o11gb/O11gb/trace
Diag Alert /u01/app/oracle/diag/rdbms/o11gb/O11gb/alert
Diag Incident /u01/app/oracle/diag/rdbms/o11gb/O11gb/incident
Diag Cdump /u01/app/oracle/diag/rdbms/o11gb/O11gb/cdump
Health Monitor /u01/app/oracle/diag/rdbms/o11gb/O11gb/hm
Default Trace File /u01/app/oracle/diag/rdbms/o11gb/O11gb/trace/O11gb_ora_16676.trc
Active Problem Count 0
Active Incident Count 0
11 rows selected.

215
ADR – V$ Diagnostic Info
11R2 – No changes (that I saw)

Optimizer Statistics &
Other Optimizer Advances
Special Thanks: Maria Colgan, Penny Avril & Debbie Migliore

217
Improved SPEED and Quality
Gathering Stats – AUTO-SAMPLING
• Manually gather stats: Impossible to find sample
size that works for ALL tables - need COMPUTE
• Especially hard to find a good sample size when the
data distribution is very skewed.
• NEW Auto-sampling: “Discovers” the best sample
size for every table in your system for you.
– Get the Quality of a COMPUTE with SPEED of a
SAMPLE
– Oracle‟ goal is to OBSOLETE the need and use of
sampling.
– Accuracy is comparable to COMPUTE

218
Incremental Statistics Maintenance -
Stats by Partition vs. table
• In 10g, if you gather stats on one partition after a bulk
load it causes a full scan of all partitions to gather global
table statistics with is extremely time consuming
• In 10g, you have to manual copy statistics to new partition
• In 11g Gather stats for TOUCHED PARTITIONS only!
• Table stats are refreshed WITHOUT scanning the
un-touched partitions.

219
Manage New Statistics
Gather Stats but make PENDING
• Currently DBAs are scared to gather stats on a table that
is changing for fear of unpredictable execution plans.
• You have to „FREEZE‟ critical plans or stats.
• In 11g, gather stats and save as PENDING.
• Verify the new stats won‟t adversely affect things by
checking them with a single user using an alter session or
try them out on a different system.
• When everything looks good – then, PUBLISH them
for all to use!

220
Gather Stats but make them PENDING
select dbms_stats.get_prefs('PUBLISH', 'SH', 'CUST') publish from dual;
PUBLISH
--------------------
TRUE
exec dbms_stats.set_table_prefs('SH', 'CUST', 'PUBLISH', 'false');
PL/SQL procedure successfully completed.
select dbms_stats.get_prefs('PUBLISH', 'SH', 'CUST') publish from dual;
PUBLISH
--------------------
FALSE

221
Gather Stats but make them PENDING
select table_name, last_analyzed analyze_time, num_rows, blocks, avg_row_len
from user_tables
where table_name = 'CUST';
TABLE_NAME ANALYZE_T NUM_ROWS BLOCKS AVG_ROW_LEN
---------- --------- ---------- ---------- -----------
CUST
execute dbms_stats.gather_table_stats('SH', 'CUST');
from user_tables
---------- --------- ---------- ---------- -----------
CUST

222
PUBLISH Stats after Testing Complete
alter session set optimizer_use_pending_statistics = true;
(Then run your query – If ready/better – publish the new stats)
exec dbms_stats.publish_pending_stats('SH', 'CUST');
from user_tables
---------- --------- ---------- ---------- -----------
CUST 13-OCT-07 55500 1485 180
exec dbms_stats.delete_table_stats('SH', 'CUST'); <to delete>

Extended Optimizer Statistics:
New Multi-Column Statistics
• Corporate data often has correlations between different
columns of a table. For example:
– A job title is correlated to the salary.
– The season affects the sold amounts of items such as swim suits
sell more in the summer and snow shoes sell more in the winter.
– The make of a car and color are often used together but are not
really correlated well so the filter doesn‟t reduce the result set.
• Optimizer has to estimate the correct cardinality
– Will the additional column condition reduce the result set or
not? Should it be used.
• Oracle calculates correlated statistics so the optimizer will
make great decisions. Single column statistics and
histograms are not enough!

Example
SELECT make, price, color
FROM cars_dot_com
WHERE make = ‘CORVETTE’;
SILVER
50,000
CORVETTE
BLACK
60,000
CORVETTE
RED
40,000
CORVETTE
• Three records selected.
• Single column statistics are
accurate
SLIVER
45,000
JEEP
BLACK
35,000
JEEP
RED
90,000
CADILLAC
SILVER
50,000
CORVETTE
BLACK
60,000
CORVETTE
RED
40,000
CORVETTE
Color
Price
Make

Example, cont.
FROM cars_dot_com
WHERE make = ‘CORVETTE’
AND COLOR = 'RED‘;
• One record selected.
• No correlated columns
• Additional predicate
reduces result set
• Single column statistics
are STILL sufficient
RED
40,000
CORVETTE
SLIVER
45,000
JEEP
BLACK
35,000
JEEP
RED
90,000
CADILLAC
SILVER
50,000
CORVETTE
BLACK
60,000
CORVETTE
RED
40,000
CORVETTE
Color
Price
Make

Example, cont.
• Three records selected.
• Correlated columns
• Additional predicate has no
effect
• Single column statistics
are NOT sufficient
• Must use ‘=‘ and not < or >
SLIVER
50,000
CORVETTE
BLACK
50,000
CORVETTE
RED
50,000
CORVETTE
SLIVER
45,000
JEEP
BLACK
35,000
JEEP
RED
90,000
CADILLAC
SILVER
50,000
CORVETTE
BLACK
50,000
CORVETTE
RED
50,000
CORVETTE
Color
Price
Make
FROM cars_dot_com
WHERE make = ‘CORVETTE’
AND PRICE = 50000;

227
Manage New Statistics – FYI Only
EXTENDED Statistic Group
• Provides a way to collect stats on a group of columns
• Full integration into existing statistics framework
– Automatically maintained with column statistics
– Instantaneous and transparent benefit for any application
• Accurate cardinalities for inter-related columns
– Multiple predicates on the same table are estimated
correctly

228
After normal Statistics Creation
select column_name, num_distinct, histogram
from user_tab_col_statistics where table_name = 'CUSTOMERS„;
COLUMN_NAME NUM_DISTINCT HISTOGRAM
------------------------------ ------------ ---------------
CUST_VALID 2 NONE
COUNTRY_ID 19 FREQUENCY
CUST_STATE_PROVINCE 145 NONE
CUST_CITY_ID 620 HEIGHT BALANCED
CUST_CITY 620 NONE
CUST_LAST_NAME 908 NONE
CUST_FIRST_NAME 1300 NONE
CUST_ID 55500 NONE
...
23 rows selected.

229
Create EXTENDED Statistic Group
• Now lets create the extended statistics group & re-gather statistics on the
CUSTOMER table (query user_tab_col_statistics to see new column):
select dbms_stats.create_extended_stats('SH','CUSTOMERS', '(country_id,
cust_state_province)') from dual;
DBMS_STATS.CREATE_EXTENDED_STATS('SH','CUSTOMERS','(CO
--------------------------------------------------------------------------------
SYS_STUJGVLRVH5USVDU$XNV4_IR#4
exec dbms_stats.gather_table_stats('SH','CUSTOMERS', method_opt =>
'for all columns size skewonly');

230
Now there are Extended Statistics
from user_tab_col_statistics where table_name = 'CUSTOMERS';
------------------------------ ------------ ---------------
SYS_STUJGVLRVH5USVDU$XNV4_IR#4 145 FREQUENCY
CUST_VALID 2 FREQUENCY
CUST_STATE_PROVINCE 145 FREQUENCY
CUST_CITY 620 HEIGHT BALANCED
CUST_LAST_NAME 908 HEIGHT BALANCED
CUST_FIRST_NAME 1300 HEIGHT BALANCED
CUST_ID 55500 HEIGHT BALANCED
...
24 rows selected.

231
DROP Extended Statistics
exec dbms_stats.drop_extended_stats('SH', 'CUSTOMERS', '(country_id, cust_state_province)');
from user_tab_col_statistics where table_name = 'CUSTOMERS„;
------------------------------ ------------ ---------------
CUST_VALID 2 NONE
CUST_STATE_PROVINCE 145 NONE
CUST_CITY 620 NONE
CUST_LAST_NAME 908 NONE
CUST_FIRST_NAME 1300 NONE
CUST_ID 55500 NONE
...
23 rows selected.

232
Adaptive Cursor Sharing
• The optimizer peeks at user-defined bind values during plan
selection on the hard parse.
• Initial value of the binds determines the plan for all future
binds (hopefully the first peek covers most queries)
• Same execution plan shared regardless of future bind values
• One plan is not always appropriate for all bind values for a
given SQL statement
– Where job= „PRESIDENT‟ (use an index – only one row)
– Where job = „OPERATOR‟ (don‟t use an index – 90% of the
table)
• If Oracle “peeks” and sees the President, it will use the
index. Future queries also use the index without peeking
after that (bad for the OPERATOR query).

233
Bind Peeking – Pre-11g
• If you need to tune a query that you suspect has issues related to
bind peeking, use v$sql_plan or tkprof output using different values
for bind variables and compare execution plans in both cases.
• If you wish to deactivate bind peeking you can set:
alter system set "_OPTIM_PEEK_USER_BINDS"=FALSE;
Note: When running tkprof "explain=username/password" argument should
NOT be used. That will cause tkprof to issue an explain plan whose output
could differ from the execution plan info inside the raw 10046/sql_trace file.

Consider a Telephone Company…
OPERATOR
7782
CLARK
OPERATOR
7788
SCOTT
PRESIDENT
8739
KING
OPERATOR
7521
WARD
OPERATOR
7499
ALLEN
OPERATOR
6973
SMITH
Job
Empno
Ename
SELECT Ename, Empno, Job
FROM Emp
WHERE Job = :B1
Value of B1 = „OPERATOR‟;
• If ‘OPERATOR’ is the bind value at
hard parse, most records will be
selected. Execution plan will be a
full table scan
• If ‘PRESIDENT’ is the bind value
at hard parse, few records will be
selected. Execution plan will be an
index search
OPERATOR
7782
CLARK
OPERATOR
7788
SCOTT
OPERATOR
7521
WARD
OPERATOR
7499
ALLEN
OPERATOR
6973
SMITH
Job
Empno
Ename

235
Adaptive Cursor Sharing
Solution:
• In 11g, Oracle uses bind-aware cursor matching.
• Share the plan when binds values are “equivalent”
– Plans are marked with selectivity range
– If current bind values fall within range they use the same
plan
• Create a new plan if binds are not equivalent
– Generating a new plan with a different selectivity range

236
Bind Peeking
Cursor Sharing (cs) Statistics
select sql_id, peeked, executions, rows_processed, cpu_time
from v$sql_cs_statistics; (using the peeked value on the 2nd+ execution)
SQL_ID P EXECUTIONS ROWS_PROCESSED CPU_TIME
------------- - ---------- -------------- ----------
5wfj3qs71nd7m Y 3 1 0
2rad83pp613m1 Y 3 3 0
dr78c03uv97bp N 1 3 0
dr78c03uv97bp N 1 3 0
dr78c03uv97bp Y 1 3 0
9qv6tq9ag5b80 Y 3 3 0
a2k4qkh681fzx Y 3 2 0
413zr99jf9h72 N 1 1 0
413zr99jf9h72 N 1 1 0
413zr99jf9h72 Y 1 1 0
fd69nfzww1mhm Y 6 0 0

237
Bind Peeking – V$SQL
select sql_id, executions, is_bind_sensitive, is_bind_aware
from v$sql;
SQL_ID EXECUTIONS I I
------------- ---------- - -
9ugwm6xmvw06u 11 Y N
bdfrydpbzw07g 11 Y N
57pfs5p8xc07w 20 N N
...
• is_bind_sensitive – If „Y‟, then Oracle is using multiple plans
depending on bind variable.
• is_bind_aware – Oracle knows that the different data patterns
may result depending on bind value. Oracle switches to a
bind-aware cursor and may hard parse the statement.

238
Host and
Hardware
Enterprise Manager for the Grid
Database
Oracle9iAS
Storage
Network and
Load Balancer
Applications
Administration
Monitoring
Provisioning
Security
Enterprise
Manager

239
Performance Manager : Back in Time!

240
Grid Control – 10gR2;
Many more Options!

244
Security Enhancements
• 11g is more restrictive
– Password lock time (1), password grace time (7) and password life
time (180) all more restrictive; Failed login attempts stays the
same (10).
– Passwords will be case sensitive now! (on by default)
– Enhanced hashing algorithm for passwords / DES still available.
– Strong passwords (set via password complexity verification in EM
or SQL):
• Minimum 8 characters
• At least one letter and one digit
• Not servername or servername(1-100)
• Not a common password (i.e. welcome1)
• Must differ from previous password by 3 characters minimum

245
Security Enhancements
AUDIT_TRAIL=DB (default)
• Audit Trail is ON by default (was off in 10g),
• AUDIT_TRAIL=DB is now the default.
• Things that will be audited by default include:
– CREATE USER, CREATE SESSION, CREATE ANY TABLLE, CREATE
ANY PROCEDURE, CREATE ANY JOB, CREATE EXTERNAL JOB,
CREATE ANY LIBRARY, CREATE PUBLIC DB LINK
– ALTER USER, ALTER ANY TABLE, ALTER ANY PROCEDURE, ALTER
PROFILE, ALTER DATABASE, ALTER SYSTEM, AUDIT SYSTEM
– DROP USER, DROP ANY TABLE, DROP ANY PROCEDURE, DROP
PROFILE
– GRANT ANY PRIVILEGE, GRANT ANY OBJECT PRIVILEGE
– EXEMPT ACCESS POLICY
– AUDIT SYSTEM
• Cost of Auditing improved to be 1-2% cost on TPCC benchmark.

246
All the Rest worth noting…
• SEC_CASE_SENSITIVE_LOGON=FALSE
• CONNECT Role only Create Session (vs. Tbl/View…)
• Consider: _NEW_INITIAL_JOIN_ORDERS=FALSE
(CBO more join orders – higher parse times possible)
• GATHER_STATS_JOB on for all DML:
DBMS_STATS.LOCK_TABLE_STATS(„SH‟,‟T1‟);
• Auto PROFILES if 3x better; Oracle Always Tuning…
• Statspack STILL works in 11g
• Real Time stats generated for high cpu queries – Careful!
• Generate System Stats on migrate: Tune / 11g Parameters

247
Oracle Audit Vault
Oracle Database Vault
DB Security Evaluation #19
Transparent Data Encryption
EM Configuration Scanning
Fine Grained Auditing (9i)
Secure application roles
Client Identifier / Identity propagation
Oracle Label Security (2000)
Proxy authentication
Enterprise User Security
Global roles
Virtual Private Database (8i)
Database Encryption API
Strong authentication (PKI, Kerberos, RADIUS)
Native Network Encryption (Oracle7)
Database Auditing
Government customer
Oracle Database Security
Built over MANY years...
2007
1977

248
Other 11gR2 Features
• Grid Plug and Play!!
• Oracle Restart – DB, ASM, Listener after restart of software/hardware
• Out of Place Upgrades (zero downtime for patching)
• In Memory Parallel Execution & Auto Degree of Parallelism (DOP)
• Enterprise Manager for Provisioning, Clusterware, GPnP, Restart
• Universal installer (Remove RAC, de-install, downgrades, patches,restarts)
• ASM FS (file system) snapshots – 64 images – backup/reco/data mining!
• Intelligent data placement on fast tracks
• Flashback Data Archive support for DDL
• Instance caging – allocate CPU usage to instances (CPU_COUNT)
• Compare SQL Tuning sets to each other
• Tuning Advisor can use auto DOP, searches historical performance, transport back
to 10gR2 or later for testing.
• Virtual Columns can be in PK/FK of reference partition table
• Stored outline migration to SQL Plan Management (SPM)
• Automatic Block Repair

249
Oracle Upgrade Case Studies
(Thanks Mike Dietrich, Carol Tagliaferri, Roy Swonger:
11g Upgrade Paper – Oracle Germany)
• University with about 20,000 users on Sun Solaris
– Moved 10 databases from 9.2.0.8 to 11.1.0.6
– Used SQL Tuning Advisor and SQL Performance Analyzer (SPA) to fix 94 queries
– Also moved to RAC, ASM & Data Guard
– 30% more logins and yet Response Time is 50% LOWER!
• International Customer with 400+ databases on IBM AIX & EMC DMX disks
– Moved from 9.2.0.8 to 11.1.0.6 – 54% slower
– Used SPM, SPA, DB Replay to tune things…
– Changed parameters to 11g – 15% improvement
– Gathered system stats – 7% improvement
– Used SPA – 18% improvement
– Turn on SQL Profiling (SPM) – 8% improvement
– 11g is now 11% FASTER than 9.2.0.8
• Data Warehouse customer on RH Linux 64-Bit
– Moving from 10.2 to 11.1.0.7 with 50 databases each at around 10T
– Over 200,000 partitions in the database
– Silent Upgrade of 50 other DWHS‟s unattended using DBUA silent mode

250
Helpful V$/X$ Queries
(FYI Only)

251
V$ Views over the years
Version V$ Views X$ Tables
6 23 ? (35)
7 72 126
8.0 132 200
8.1 185 271
9.0 227 352
9.2 259 394
10.1.0.2 340 (+31%) 543 (+38%)
10.2.0.1 396 613
11.1.0.6.0 484 (+23%) 798 (+30%)
11.2.0.1.0 496 (+25%) 945 (+54%)

252
Listing of V$ Views
select name
from v$fixed_table
where name like 'GV%'
order by name;
NAME
---------------------------------
GV$ACCESS
GV$ACTIVE_INSTANCES
GV$ACTIVE_SESS_POOL_MTH
GV$AQ1
GV$ARCHIVE…

253
Need GV$ - Instance ID
select (1 - (sum(decode(name, 'physical reads',value,0)) /
(sum(decode(name, 'db block gets',value,0)) +
sum(decode(name, 'consistent gets',value,0))))) * 100 “Hit Ratio"
from v$sysstat;
Hit Ratio
------------------
90.5817699

254
Need GV$ - Instance ID
select inst_id, (1 - (sum(decode(name, 'physical reads',value,0)) /
(sum(decode(name, 'db block gets',value,0)) +
sum(decode(name, 'consistent gets',value,0))))) * 100 “Hit Ratio"
from gv$sysstat
group by inst_id;
INST_ID Hit Ratio
-------------- ------------------
1 90.5817699
2 96.2034537

255
X$ used to create V$
select *
from v$fixed_view_definition
where view_name = 'GV$INDEXED_FIXED_COLUMN‟;
VIEW_NAME VIEW_DEFINITION
gv$indexed_fixed_column select c.inst_id, kqftanam,
kqfcoidx, kqfconam,kqfcoipo
from X$kqfco c, X$kqfta t
where t.indx = c.kqfcotab
and kqfcoidx != 0

256
Listing of X$ Tables
select name
from v$fixed_table
where name like 'X%'
order by name;
NAME
---------------------------------
X$ACTIVECKPT
X$BH
X$BUFQM
X$CKPTBUF
X$CLASS_STAT…

257
Listing of X$ Indexes
(498 in 11.1.0.3.0, 419 in 10g; 326 in 9i)
select table_name, index_number, column_name
from gv$indexed_fixed_column
order by table_name, index_number, column_name,
column_position;
TABLE_NAME INDEX_NUMBER COLUMN_NAME
------------------------------ ------------------------------ ------------------------------
X$CLASS_STAT 1 ADDR
X$CLASS_STAT 2 INDX
X$DUAL 1 ADDR
X$DUAL 2 INDX …

258
V$ - System Information
select * from v$version;
BANNER
---------------------------------------------------------------------------------------------
Oracle Database 11g Enterprise Edition Release 11.1.0.3.0 - Beta
PL/SQL Release 11.1.0.3.0 - Beta
CORE 11.1.0.3.0 Beta
TNS for Linux: Version 11.1.0.3.0 - Beta
NLSRTL Version 11.1.0.3.0 - Beta

259
V$ - System Information
select *
from v$option;
PARAMETER VALUE
----------------------------------- -------------
Partitioning TRUE
Objects TRUE
Real Application Clusters FALSE
Advanced Replication TRUE
Bit-Mapped Indexes TRUE
…

260
V$ - V$SESSION_WAIT
(waiting right now)
select event, sum(decode(wait_time,0,1,0)) “Waiting Now",
sum(decode(wait_time,0,0,1)) “Previous Waits",
count(*) “Total”
from v$session_wait
group by event
order by count(*);
WAIT_TIME = 0 means that it‟s waiting
WAIT_TIME > 0 means that it previously waited this many ms

261
V$ - V$SESSION_WAIT
EVENT Waiting Now Previous Waits Total
--------------------------- ------------------- --------------------- ------------
db file sequential read 0 1 1
db file scattered read 2 0 2
latch free 0 1 1
enqueue 2 0 2
SQL*Net message from client 0 254 480
…

262
V$SESSION_WAIT
Current Specific waits
Buffer Busy Waits or Write Complete Waits Events:
SELECT /*+ ordered */ sid, event, owner, segment_name,
segment_type,p1,p2,p3
FROM v$session_wait sw, dba_extents de
WHERE de.file_id = sw.p1
AND sw.p2 between de.block_id
and de.block_id+de.blocks – 1
AND (event = 'buffer busy waits'
OR event = 'write complete waits')
AND p1 IS NOT null
ORDER BY event,sid;

263
V$EVENT_NAME
Finding P1, P2, P3
col name for a20
col p1 for a10
col p2 for a10
col p3 for a10
select event#,name,parameter1 p1,parameter2 p2,parameter3 p3
from v$event_name
where name in ('buffer busy waits', 'write complete waits')
EVENT# NAME P1 P2 P3
------------- -------------------- ---------- ---------- ----------
74 write complete waits file# block#
76 buffer busy waits file# block# class#

264
V$ - V$SESSION_WAIT_HISTORY
(Last 10 waits for session)
Buffer Busy Waits or Write Complete Waits Events:
SELECT /*+ ordered */ sid, event, owner, segment_name,
segment_type,p1,p2,p3
FROM v$session_wait_history sw, dba_extents de
WHERE de.file_id = sw.p1
AND sw.p2 between de.block_id
and de.block_id+de.blocks – 1
AND (event = 'buffer busy waits'
OR event = 'write complete waits')
AND p1 IS NOT null
ORDER BY event,sid;

265
Great V$ - V$SESSION_EVENT
(waiting since the session started)
select sid, event, total_waits, time_waited, event_id
from v$session_event
where time_waited > 0
order by time_waited;
SID EVENT TOTAL_WAITS TIME_WAITED
---------- ------------------------------ ----------- -----------
159 process startup 2 1
167 latch: redo allocation 4 1
168 log buffer space 2 3
166 control file single write 5 4
…

266
V$ - V$SESSION_WAIT_CLASS
(session waits by WAIT CLASS)
select wait_class, total_waits
from v$system_wait_class;
WAIT_CLASS TOTAL_WAITS
-------------------- -----------
Other 4180
Application 45269
Configuration 297
Concurrency 25467
Commit 54805
Idle 6925277
Network 1859009
User I/O 809979
System I/O 1103539
Scheduler 10276

267
Great V$ - V$SYSTEM_EVENT
(waits since the instance started)
EVENT TOTAL_WAITS TIME_WAITED
------------------------------ ----------- -----------
latch: session allocation 5644 1
latch: redo allocation 4 1
latch: cache buffers chains 4 3
enq: TX - index contention 1 3
direct path write temp 57 6
row cache lock 1 7 …
select event, total_waits, time_waited
from v$system_event
where time_waited > 0
order by time_waited;

268
“cursor: pin S wait on X” WAITEVENT
• P1 = The idn which is the HASH_VALUE of SQL statement we
are waiting on. You can query V$SQL or V$SQLAREA to get the
actual SQL_TEXT.
select sql_text
from v$sqlarea
where hash_value=&&P1;
• P2 is the mutex value, which is the session id (higher bits) and
reference count (which is 0 if you are sleeping/waiting for an X mode
holder).
select decode(trunc(&&P2/4294967296), 0, trunc(&&P2/65536),
trunc(&&P2/4294967296)) SID_HOLDING_THE_MUTEX
from dual;

269
Find the mutex blocker…
• In 11g, you can also go to V$SESSION to see the
blocker:
select action, blocking_session, blocking_session_status,
sql_id
from v$session
where sid = &SID_HOLDING_THE_MUTEX;
(Get SID_HOLDING_THE_MUTEX from query above)

270
Also use V$MUTEX_SLEEP and
V$MUTEX_SLEEP_HISTORY
select mutex_type, count(*), sum(sleeps)
from v$mutex_sleep
group by mutex_type;
MUTEX_TYPE COUNT(*) SUM(SLEEPS)
-------------------------------- ---------- -----------
Library Cache 8 3891
Cursor Pin 1 122
select mutex_type, count(*), sum(sleeps)
from v$mutex_sleep_history
group by mutex_type;
MUTEX_TYPE COUNT(*) SUM(SLEEPS)
-------------------------------- ---------- -----------
Library Cache 18 3891
Cursor Pin 25 117

271
V$ - Top 10 as % of All
select sum(pct_bufgets) percent
from (select rank() over ( order by buffer_gets desc ) as rank_bufgets,
to_char(100 * ratio_to_report(buffer_gets) over (), '999.99')
pct_bufgets
from v$sqlarea )
where rank_bufgets < 11;
PERCENT
------------
97.07

272
V$ - What Users are doing…
select a.sid, a.username, s.sql_text
from v$session a, v$sqltext s
where a.sql_address = s.address
and a.sql_hash_value = s.hash_value
order by a.username, a.sid, s.piece;
SID USERNAME SQL_TEXT
------ ------------------- ------------------------------------
11 PLSQL_USER update s_employee set salary = 10000
9 SYS select a.sid, a.username, s.sql_text
9 SYS from v$session a, v$sqltext
9 SYS where a.sql_address = s.address
(…partial output listing)

273
Great V$ - V$SEGMENT_STATISTICS
select object_name, statistic_name, value
from v$segment_statistics
where value > 100000
order by value;
OBJECT_NAME STATISTIC_NAME VALUE
---------------------------- ------------------------------------- ---------------
ORDERS space allocated 96551
ORDERS space allocated 134181
ORDERS logical reads 140976
ORDER_LINES db block changes 183600

274
AWR – Segments by
Buffer Busy Waits

275
AWR – Segments by Logical Reads

276
(Thanks: Veljko Lavrnic (VL), Scott Martin, Tirth (tink), Andy, Nitin, Kevin)
Oracle Internals at the Block Level ;
Beginners close your eyes!
If time permits - Ohio 2011

277
Current & CR Versions
• Buffer hash table x$bh holds headers (hash chain protected
by a CBC latch) point to db_block buffers in memory.
• For a given block - Only one block is CURRENT and no
more than 5 other CR versions of the block (as of V9).
• For DML, you need the CURRENT version.
• For query, you can use the CURRENT version if not being
used and/or build a CONSISTENT READ (CR) version
by applying and undo needed. This may include reading
the ITL, mapping to the UNDO HEADER, but the ITL
also maps directly to the UNDO BLOCK and applying the
UNDO to get the correct CR version that you need.
• Links for the LRU & LRU-W (working set used for buffer
replacement) are maintained in the buffer headers.

278
Biggest Problems
• The SQL, of course… especially reads of full indexes, full
table scans, bad table joins and others.
• Hot blocks… hot blocks can cause latching issues. Bad
SQL or bad indexes causes hot blocks (scanning through
the same large index). Improved in 10g (shared latches).
• Not enough freelists or not using ASSM.
• Not enough initrans for multiple DML to the same block
(pctfree not high enough to auto-generate). Or too many
(each ITL costs 24 bytes).
• Slow I/O subsystem or poor disk caching or not enough
paths and readers/writers colliding.
• Not on latest version so can‟t use great new features!

280
Last Resort - Block Dumps
SQL> desc emp1
Name Null? Type
---------------------------------------- -------- ----------------------------
EMPNO NUMBER(4)
ENAME VARCHAR2(10)
JOB VARCHAR2(9)
MGR NUMBER(4)
HIREDATE DATE
SAL NUMBER(7,2)
COMM NUMBER(7,2)
DEPTNO NUMBER(2)

281
select *
from emp
where ename = 'MILLER';
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- -------------- ------------ --------- ------------------- -------- ----------- --------------
7934 MILLER CLERK 7782 23-JAN-82 1300 10

282
select file_id, block_id, blocks
from dba_extents
where segment_name = 'EMP'
and owner = 'SCOTT';
FILE_ID BLOCK_ID BLOCKS
----------- --------------- ------------
1 50465 3

283
ALTER SYSTEM DUMP DATAFILE 5 BLOCK 50465
/
/
/
Or…
ALTER SYSTEM DUMP DATAFILE 5 BLOCK MIN
50465 BLOCK MAX 50467;
(Puts output in user_dump_dest)

284
Block Dump… Getting the
block number
select rowid,empno,
dbms_rowid.rowid_relative_fno(rowid) fileno,
dbms_rowid.rowid_block_number(rowid) blockno,
dbms_rowid.rowid_row_number(rowid) rowno, rownum,
rpad(to_char(dbms_rowid.rowid_block_number(rowid), 'FM0xxxxxxx') || '.' ||
to_char(dbms_rowid.rowid_row_number (rowid), 'FM0xxx' ) || '.' ||
to_char(dbms_rowid.rowid_relative_fno(rowid), 'FM0xxx' ), 18) myrid
from emp1;
ROWID EMPNO FILENO BLOCKNO ROWNO ROWNUM
------------------ ---------- ---------- ---------- ---------- ----------
MYRID
------------------
AAAMfcAABAAAN0KAAA 7369 1 56586 0 1
0000dd0a.0000.0001
AAAMfcAABAAAN0KAAB 7499 1 56586 1 2
0000dd0a.0001.0001
AAAMfcAABAAAN0KAAC 7521 1 56586 2 3
0000dd0a.0002.0001

286
Block Dumps – Top Section
*** 2005-04-08 23:18:49.226
Start dump data blocks tsn: 0 file#: 1 minblk 56650 maxblk 56650
buffer tsn: 0 rdba: 0x0040dd4a (1/56650)
scn: 0x0000.003dfa58 seq: 0x01 flg: 0x00 tail: 0xfa580601
frmt: 0x02 chkval: 0x0000 type: 0x06=trans data
Block header dump: 0x0040dd4a
Object id on Block? Y
seg/obj: 0xce1c csc: 0x00.3dfa58 itc: 2 flg: O typ: 1 - DATA
fsl: 0 fnx: 0x0 ver: 0x01
Scn block
was last
changed at
ACTUAL
Database Block
Address
Changes to
block w/i scn Blocks that were
dumped
#ITL’s (see next slide)

287
Itl Xid Uba Flag Lck Scn/Fsc
0x01 0x0004.010.00000fba 0x0080003d.08b5.10 ---- 4 fsc 0x009d.00000000
0x02 0x0004.016.00000fae 0x008000cc.08af.34 C--- 0 scn 0x0000.003deb5b
ITL – 2 Interested
Transaction Lists
Transaction ID
Undo#.slot#.wrap#
(Undo#,slot#,seq#)
UBA:
File.block(Undo dba).sequence.record
Undo block address where last change
is recorded.
Rows Locked:
4 rows deleted
for this xid in
this block.

288
0x01 0x0004.010.00000fba 0x0080003d.08b5.10 ---- 4 fsc 0x009d.00000000
Flag: No flag set then it’s uncommitted (----)/(CBUT)
C---=Committed
-B-- = The UBA contains undo for this itl
--U-= Committed (scn is upper bound)…used by fast commits &
delayed block cleanout has not occurred
---T = Transaction active at block cleanout SCN
C-U- = Block cleaned by delayed block cleanout, and the
rollback segment info overwritten. The scn will show the lowest
scn that could be regenerated by the rollback segment.

289
0x01 0x0004.010.00000fba 0x0080003d.08b5.10 ---- 4 fsc 0x009d.00000000
Scn/fsc:
Fsc= Free Space Credit = 9d (hex) = 157 (decimal) bytes
Scn = System change (commit) number

290
data_block_dump,data header at 0x5a1125c
===============
tsiz: 0x1fa0 Total Size = 1fa0 = 8096
1 (hex) = 1 (decimal) x 16x16x16 = 4096
F (hex) = 15 (decimal) x 16x16 = 3840
A (hex) = 10 (decimal) x 16 = 160
0 (hex) = 0 (decimal) x 1 = 0
4096 + 3840 + 160 + 0 = 8096 (about 8K)
hsiz: 0x2e Header Size = 2e = 46 bytes
pbl: 0x05a1125c Pointer to block buffer holding the block
bdba: 0x0040dd4a (Relative) database block address

291
Block Dumps – Data Section
ntab=1 Number of tables= 1 (2+ for clusters)
nrow=14 Number of rows = 14
frre=-1 First free row index entry – 1 (add 1)
fsbo=0x2e Free space begin offset
fseo=0x18fb Free space end offset
avsp=0x1d3b Available block space = 1d3b = 7483
tosp=0x1de0 Space avail. post commit=1de0=7648
0xe:pti[0] nrow=14 offs=0 Table Info
0x12:pri[0] offs=0x18fb (6395)Row Info Record 0
0x14:pri[1] offs=0x1921 (6433)Row Info Record 1
… Row Info Records 2–13
block_row_dump: Row Data is Next!

293
Block Dumps – output from udump
tab 0, row 13, @0x1b0b
tl: 39 fb: --H-FL-- lb: 0x0 cc: 8
col 0: [ 3] c2 50 23
col 1: [ 6] 4d 49 4c 4c 45 52
col 2: [ 5] 43 4c 45 52 4b
col 3: [ 3] c2 4e 53
col 4: [ 7] 77 b6 01 17 01 01 01
col 5: [ 2] c2 0e
col 6: *NULL*
col 7: [ 2] c1 0b…

294
DUMP OUTPUT:
tl: 39 fb: --H-FL-- lb: 0x0 cc: 8 (row header)
Table = this data is for table 0
Row 13 = 14th Row (0-13 total rows)
Offset: 1b0b (in Hex) – Offset from header
tl: Total bytes of row plus the header = 39

295
DUMP OUTPUT:
tl: 39 fb: --H-FL-- lb: 0x0 cc: 8
fb: --H-FL-- = flag byte; ( -KCHDFLPN)
H = Head of row piece, F = First data piece, L=Last piece
D = Deleted; P= First column continues from previous piece
(chaining) ; N= Last column continues in next piece;
K = Cluster Key; C = Cluster table member

296
DUMP OUTPUT:
tl: 39 fb: --H-FL-- lb: 0x0 cc: 8
Lb: lock byte is 1+ if this row is locked = 0 (unlocked)
cc: Column count = 8

297
Block Dumps – Deleted rows
DUMP OUTPUT (Deleted Row):
block_row_dump:
tab 0, row 0, @0x18fb
tl: 2 fb: --HDFL-- lb: 0x1
tab 0, row 1, @0x1921
tl: 2 fb: --HDFL-- lb: 0x1
Rows 1 &2 have been deleted! No row data is visible
for the columns.

298
DUMP OUTPUT - EMPNO:
col 0: [ 3] c2 50 23
Hex to Decimal: Col0 = EMPNO = 7934
50 (Hex) = 80 (Decimal) – 1 = 79
23 (Hex) = 35 (Decimal) – 1 = 34
c2: Number in the thousands (c2 is exponent)

299
DUMP OUTPUT - ENAME:
col 1: [ 6] 4d 49 4c 4c 45 52
Hex to Character: Col1 = ENAME = MILLER
4d (Hex) = M (Character)
49 (Hex) = I (Character)
4c (Hex) = L (Character)
45 (Hex) = E (Character)
52 (Hex) = R (Character)

300
DUMP OUTPUT - JOB:
col 2: [ 5] 43 4c 45 52 4b
Hex to Character: Col2 = JOB = CLERK
43 (Hex) = C (Character)
45 (Hex) = E (Character)
52 (Hex) = R (Character)
4b (Hex) = K (Character)

301
DUMP OUTPUT - MGR:
col 3: [ 3] c2 4e 53
Hex to Decimal: Col3 = MGR = 7782
4e (Hex) = 78 (Decimal) – 1 = 77
53 (Hex) = 83 (Decimal) – 1 = 82

302
DUMP OUTPUT - HIREDATE:
col 4: [ 7] 77 b6 01 17 01 01 01
Hex to Decimal: Col4 = HIREDATE = 23-JAN-82
77 (Hex) = 119 (Decimal) – 100 = 19
B6 (Hex) = 182 (Decimal) – 100 = 82
01(Hex) = 1 (Decimal) <month>
17 (Hex) = 23 (Decimal) <day>
01 01 01 (Hex) = This is the Hour, Minute, Second
(none were entered when the date was entered…default)

303
DUMP OUTPUT - SAL:
col 5: [ 2] c2 0e
Hex to Decimal: Col5 = SAL = 1300
0e (Hex) = 14 (Decimal) – 1 = 13
c2 = add two zero‟s

304
DUMP OUTPUT - COMM:
col 6: *NULL*
Hex to Decimal: Col6 = COMM = NULL
NULL = NULL

305
DUMP OUTPUT - DEPTNO:
col 7: [ 2] c1 0b
Hex to Decimal: Col7 = DEPTNO= 10
0B (Hex) = 11 (Decimal) – 1 = 10
c1 = number in the tens

306
Block Dump:
Data Section – Other Ways

307
Block Dump using SELECT dump()
select dump(ename) from emp1
where ename='MILLER„;
DUMP(ENAME)
------------------------------------------------------
Typ=1 Len=6: 77,73,76,76,69,82
Types: 1=varchar; 2=number; 12=date; 23=raw

308
Typ=1 Len=6: 77,73,76,76,69,82
Decimal to Character: ENAME = MILLER
77 (Decimal) = M (Character)
73 (Decimal) = I (Character)
76 (Decimal) = L (Character)
76 (Decimal) = L (Character)
69 (Decimal) = E (Character)
82 (Decimal) = R (Character)

309
(convert it to HEX if you want)
select dump(ename.16) from emp1
where ename='MILLER„;
DUMP(ENAME,16)
------------------------------------------------------
Typ=1 Len=6: 4d,49,4c,4c,45,52
Types: 1=varchar; 2=number; 12=date; 23=raw

310
(Can even get the ename from the HEX!)
select dump(ename,16), ename
from emp1
where dump(ename,16) like '%4d,49,4c,4c,45,52„;
DUMP(ENAME,16) ENAME
---------------------------------------------- --------------
Typ=1 Len=6: 4d,49,4c,4c,45,52 MILLER

311
Transactions
Moving through Oracle

312
Row Level Locks
(Chain
Latch)
n
1
3
Cache Buffers
Buffer
Header
Hash
Buckets
LRU Lists LRU Write List
DB
Write
1
2
3
Log
Buffers
User
Request
Shared
Pool
SGA
Buffer Cache
2
Cache Layer
2
Transaction Layer
XID 01
XID 02
2 Row 3
1 Row 1
1 Row 2
Lock Byte
ITL 1
ITL 2
XID 01 ITL 1
Itl Xid Uba Flag Lck …
0x01 0x0005.02a.0000028c 0x008000af.026b.01 ---- 1 …
usn#
0x02a 0x028c
RBU Header 5
wrap#
Data Block
slot#
XID
0X005.02a.0000028c
Transaction Identifiers

(Chain
Latch)
n
1
2
3
Buffer Headers
Buffer
Header
Hash
Buckets
Working with Hash Buckets
And Buffer Headers (not buffers)
• Users asks for a specific data
block address.
• This is hashed with a hashing
algorithm and looks in the hash
bucket that it hashes to.
• It walks the hash chain using
the cache buffers chain latch to
find the block that it needs
(curr or cr). There can be many
versions of each block
• Block maps to memory
address

314
_DB_BLOCK_HASH_BUCKETS
and hashing data block addresses
Example: _DB_BLOCK_HASH_BUCKETS
(shouldn‟t have to change this in Oracle9i or 10g)
• Buffer hash table (x$bh) has all buffer headers for all db_block buffers.
• Buffer header ties to memory base address of the buffer.
• Buckets usually set to Prime(2*db_block_buffers)
• A prime number is often used to avoid hashing anomalies
• Objects dba (class) is hashed to a hash bucket on the hash chain
• Get enough hash buckets (_db_block_hash_buckets)
• Blocks assigned to a hash bucket and onto the hash chain
• Could have multiple blocks hashed to same chain (if both hot-issues)
• Can have multiple versions of a block on same chain
• When block is replaced (based on LRU chain) new block comes in and
could be (probably will be) hashed to a different hash chain.

315
Example: Emp Table
Consider a user querying emp:
• First block of emp may go to chain #1
• Second block of emp may go to chain #55
• If second block of emp is updated and also has several
readers than we‟ll get more copies. LRBA – Lowest Redo
Block Address (last redo applied) for dirty block.
• Chain #55 may now have a current block and 2 CR
blocks all with the same dba (data block address)
• For a given block - Only one block is CURRENT and no
more than 5 other CR versions of the block (as of V9).
• All buffer headers tie to LRU, LRU-W and other LRU‟s
(many in 10g) used for buffer replacement.

316
Additional LRU‟s / Faster!!
• LRU Main block replacement list
• LRU-W Old dirty buffers and reco/temp
• LRU-P Ping Buffer list / RAC
• LRU-XO Buffers to be written for drop/truncate
• LRU-XR Buffers to be written for reuse range
• Thread CKPT Thread Checkpoint Queue
• File CKPT File Checkpoint Queue
• Reco CKPT Reco Checkpoint
• LRU-MAIN & LRU-AUX help LRU

317
Query all buffer headers (state):
col status for a6
select state,
decode(state, 0, 'FREE', /* not currently is use */
1, 'XCUR', /* held exclusive by this instance */
2, 'SCUR', /* held shared by this instance */
3, 'CR', /* only valid for consistent read */
4, 'READ', /* is being read from disk */
5, 'MREC', /* in media recovery mode */
6, 'IREC„, /* in instance(crash) recovery mode */
7, „WRITE‟, /* being written */
8, „PIN‟) status, count(*) /* pinned */
from x$bh
group by state;
STATE STATUS COUNT(*)
------------ ------------- -----------------
1 XCUR 2001 -- CURRENT (CURR)
3 CR 3 -- CONSISTENT READ (CR)

318
EMP1 is Block#: 56650
(all rows are in this block)
select rowid,empno,
from emp1;
------------------ ---------- ---------- ---------- ---------- ----------
MYRID
------------------
AAAM4cAABAAAN1KAAA 7369 1 56650 0 1
0000dd4a.0000.0001
AAAM4cAABAAAN1KAAB 7499 1 56650 1 2
0000dd4a.0001.0001
…
…
AAAM4cAABAAAN1KAAN 7934 1 56650 13 14
0000dd4a.000d.0001
14 rows selected.

319
from x$bh
order by dbablk;
---------- ---------- ---------- ---------- ---------- ---------- ----------
---------- - - - - - -
0 1 1 56650 0 35659776 4294967295
0 N N N N N N

Log
Buffers
(Chain
Latch)
n
1
2
3
Shared
Pool
SGA
Cache Buffers
Buffer Cache
Buffer
Header
Hash
Buckets
Only ONE block
on the Hash
Chain!

321
(Delete a row)
delete from emp1
where comm = 0;
one row deleted.

322
(Make some changes 2 copies)
from x$bh
order by dbablk;
---------- ---------- ---------- ---------- ---------- ---------- ----------
---------- - - - - - -
0 1 1 56650 1 8200 4294967295
0 N N N N N N
0 3 1 56650 2 524288 0
4347881 N N N N N N

Log
Buffers
(Chain
Latch)
n
1
2
3
Shared
Pool
SGA
Cache Buffers
Buffer Cache
Buffer
Header
Hash
Buckets
Hash Chain is
now TWO! One
is a CR and the
other is Current.

324
V$Transaction now has our record
(created when transactions have undo)
SELECT t.addr, t.xidusn USN, t.xidslot SLOT, t.xidsqn SQL, t.status,
t.used_ublk UBLK, t.used_urec UREC, t.log_io LOG,
t.phy_io PHY, t.cr_get, t.cr_change CR_CHA
FROM v$transaction t, v$session s
WHERE t.addr = s.taddr;
ADDR USN SLOT SQL STATUS UBLK
-------- ---------- ---------- ---------- ---------------- ----------
UREC LOG PHY CR_GET CR_CHA
---------- ---------- ---------- ---------- ----------
69E50E5C 5 42 652 ACTIVE 1
1 3 0 3 0
USN is the Undo Segment Number (rollback segment ID)
SLOT is the slot number in the rollback segment‟s transaction table.
SQN (Wrap) is the sequence number for the transaction.
USN+SLOT+SQN are the three values that uniquely identifies a transaction XID

325
UBAFIL is the file for last undo entry
UBLK is block for last undo entry (find out how many undo blocks).
UBASQN is the sequence no of the last entry.
UREC is the record number of the block( shows how many table and index
entries the transaction has inserted, updated or deleted.
If you are doing an INSERT or DELETE, then you will see that UREC is
set to <number of indexes for this table> + how many rows you
inserts/deletes. If you UPDATE a column then UREC will be set to
<number of indexes that his column belongs to> * 2 + number of
updated rows (so if the column belongs to no index, then UREC is set
to the number of rows that was updated).
If USED_UBLK and USED_UREC are decreasing each time you query,
then the transaction is rolling back. When USED_UREC zero, the
rollback is finished.
V$Transaction (FYI Only)

326
Dump the block
Dump the block
0x01 0x0005.02a.0000028c 0x008000af.02b6.01 ---- 1 fsc 0x0029.00000000
ITL – 2 Interested
Transaction Lists
Transaction ID
Undo 5 = 5 (decimal)
Slot 2a = 42 (decimal)
SEQ 28C = 652
UBA:
File.block.sequence.record
Undo block address where
last change is recorded.
The row I
deleted is still
locked; fsc is
0x29 = 41 bytes
Committed
Transaction

327
Insert in 4 other sessions & drive x$bh
up to the max of 6 versions of block
---------- ---------- ---------- ---------- ---------- ---------- ----------
---------- - - - - - -
0 3 1 56650 1 524416 0
4350120 N N N N N N
0 3 1 56650 1 524416 0
4350105 N N N N N N
365 1 1 56650 7 33562633 4350121
0 Y N N N N N
0 3 1 56650 1 524416 0
4350103 N N N N N N
0 3 1 56650 1 524416 0
4350089 N N N N N N
0 3 1 56650 1 524288 0
4350087 N N N N N N

Log
Buffers
(Chain
Latch)
n
1
2
3
Shared
Pool
SGA
Cache Buffers
Buffer Cache
Buffer
Header
Hash
Buckets
Hash Chain is
now SIX long!
Five CR and the
one Current.

329
Why only 6 versions of a Block?
select a.ksppinm, b.ksppstvl, b.ksppstdf, a.ksppdesc
from x$ksppi a, x$ksppcv b
where a.indx = b.indx
and substr(ksppinm,1,1) = '_'
and ksppinm like '%&1%'
order by ksppinm;
KSPPINM
-------------------------------------------------------------------------------
KSPPSTVL
-------------------------------------------------------------------------------
KSPPSTDF
---------
KSPPDESC
-------------------------------------------------------------------------------
_db_block_max_cr_dba
6
TRUE
Maximum Allowed Number of CR buffers per dba

330
User 1 – Updates Row# 1&2
Log
Buffers
Shared
Pool
SGA
Buffer Cache
User 3 Updates Row# 3
Cache Layer
2
Transaction Layer
XID 01
XID 02
2 Row 3
1 Row 1
1 Row 2
Lock Byte
ITL 1
ITL 2
User1
Request
User3
Request
XID 02 ITL 2
Row 3
2 Row 3
• User1 updates 2 rows with an
insert/update/delete – an ITL is
opened and xid tracks it in the data
block (lock byte is set on row).
• The xid ties to the UNDO header
block which ties to the UNDO data
block for undo.
• If user2 wants to query the row, they
create a clone and rollback the
transaction going to the undo header
and undo block.
• If user3 wants to update same row
(they wait). If user 3 wants to
update different row, then they open a
second ITL with an xid that maps to
an undo header & maps to an undo
block.

331
Create EMP2 („MILLER‟/‟ALLEN‟)
create table emp2
asselect * from emp1
where ename in ('MILLER','ALLEN');
select empno, ename, job
from emp2;
EMPNO ENAME JOB
-------------- -------------- ---------------------
7499 ALLEN SALESMAN
7934 MILLER CLERK

332
Get the Blockno for EMP2
select rowid,empno,
from emp2;
------------------ ---------- ---------- ---------- ---------- ----------
MYRID
------------------
AAANB2AABAAAOHSAAA 7499 1 57810 0 1
0000e1d2.0000.0001
AAANB2AABAAAOHSAAB 7934 1 57810 1 2
0000e1d2.0001.0001

333
Dump the EMP2 block (Partial)
Alter system dump datafile 2 block 57810;
System Altered.
*** SESSION ID:(154.36) 2005-04-25 23:24:22.865
buffer tsn: 0 rdba: 0x0040e1d2 (1/57810)
scn: 0x0000.00432370 seq: 0x02 flg: 0x04 tail: 0x23700602
frmt: 0x02 chkval: 0xb205 type: 0x06=trans data
Block header dump: 0x0040e1d2
seg/obj: 0xd076 csc: 0x00.43236f itc: 3 flg: - typ: 1 - DATA
0x01 0xffff.000.00000000 0x00000000.0000.00 C--- 0 scn 0x0000.0043236f
0x02 0x0000.000.00000000 0x00000000.0000.00 ---- 0 fsc 0x0000.00000000
0x03 0x0000.000.00000000 0x00000000.0000.00 ---- 0 fsc 0x0000.00000000

334
block_row_dump:
tab 0, row 0, @0x1f5d
tl: 43 fb: --H-FL-- lb: 0x0 cc: 8
col 0: [ 3] c2 4b 64
col 1: [ 5] 41 4c 4c 45 4e A L L E N
col 2: [ 8] 53 41 4c 45 53 4d 41 4e
col 3: [ 3] c2 4d 63
col 4: [ 7] 77 b5 02 14 01 01 01
col 5: [ 2] c2 11
col 6: [ 2] c2 04
col 7: [ 2] c1 1f

335
tab 0, row 1, @0x1f36
tl: 39 fb: --H-FL-- lb: 0x0 cc: 8
col 0: [ 3] c2 50 23
col 1: [ 6] 4d 49 4c 4c 45 52 M I L L E R
col 2: [ 5] 43 4c 45 52 4b
col 3: [ 3] c2 4e 53
col 4: [ 7] 77 b6 01 17 01 01 01
col 5: [ 2] c2 0e
col 6: *NULL*
col 7: [ 2] c1 0b
end_of_block_dump

336
Update „MILLER‟ to „SMALL‟
update emp2
set ename = „SMALL‟
where ename = 'MILLER';
from emp2;
EMPNO ENAME JOB
-------------- -------------- ---------------------
7499 ALLEN SALESMAN
7934 SMALL CLERK

337
System Altered.
buffer tsn: 0 rdba: 0x0040e1d2 (1/57810)
scn: 0x0000.00432794 seq: 0x05 flg: 0x00 tail: 0x27940605
frmt: 0x02 chkval: 0x0000 type: 0x06=trans data
Block header dump: 0x0040e1d2
seg/obj: 0xd076 csc: 0x00.43236f itc: 3 flg: O typ: 1 - DATA
0x01 0xffff.000.00000000 0x00000000.0000.00 C--- 0 scn 0x0000.0043236f
0x02 0x0004.02a.000012ff 0x00800353.0a9e.07 ---- 1 fsc 0x0001.00000000
0x03 0x0000.000.00000000 0x00000000.0000.00 ---- 0 fsc 0x0000.00000000

338
Here is the ITL with our Transaction
0x02 0x0004.02a.000012ff 0x00800353.0a9e.07 ---- 1 fsc 0x0001.00000000
Transaction ID
Seq 12ff = 4863 (decimal)
UBA:
The row that
was updated is
still locked; fsc
is 0x1 = 1 bytes

339
Find the Segment & Location of RBS
select segment_name
from dba_rollback_segs
where segment_id = 4;
SEGMENT_NAME
------------------------------
_SYSSMU4$
select header_file, header_block
from dba_segments
where segment_name = „_SYSSMU4$‟;
HEADER_FILE HEADER_BLOCK
-------------------------- ------------------------------
2 57

340
Dump the UNDO Header (Partial)
Transaction Table (last modified blk)!
System Altered.
TRN TBL::
index state cflags wrap# uel scn dba (uba)
parent-xid nub stmt_num cmt
-----------------------------------------------------------------------
-------------------------
0x00 9 0x00 0x12ff 0x0001 0x0000.00432687 0x0080034e
0x0000.000.00000000 0x00000001 0x00000000 1114490213
0x01 9 0x00 0x12ff 0x0002 0x0000.0043269d 0x0080034e
0x0000.000.00000000 0x00000001 0x00000000 1114490272
...
0x29 9 0x00 0x12ff 0x0028 0x0000.004327a4 0x00800353
0x0000.000.00000000 0x00000001 0x00000000 1114490829
0x2a 10 0x80 0x12ff 0x0002 0x0000.00432795 0x00800353
0x0000.000.00000000 0x00000001 0x00000000 0
…
End dump data blocks tsn: 1 file#: 2 minblk 57 maxblk 57

341
Dump the UNDO Header (Partial)
Transaction Table!
index state cflags wrap# uel scn dba
---------------------------------------------------------------
0x2a 10 0x80 0x12ff 0x0002 0x0000.00432795 0x00800353
State:
State 10 is Uncommitted
Wrap/Seq:
The Wrap is 12ff
The scn for
uncommitted
(ours is) or
committed
transactions
Slot:
The Slot was 2a
which was the
42nd in the list.
The dba of the
undo which we
need to look in.

342
The ITL again… (fyi to see UBA)
0x02 0x0004.02a.000012ff 0x00800353.0a9e.07 ---- 1 fsc 0x0001.00000000
Transaction ID
Seq 12ff = 4863 (decimal)
UBA:
The row that
was updated is
still locked; fsc
is 0x1 = 1 bytes

343
Find the Segment & Location of RBS
SELECT DBMS_UTILITY.DATA_BLOCK_ADDRESS_FILE(
TO_NUMBER('00800353','XXXXXXXX')) UFILE
FROM DUAL;
UFILE
-----------
2
SELECT DBMS_UTILITY.DATA_BLOCK_ADDRESS_BLOCK(
TO_NUMBER('00800353','XXXXXXXX')) BLOCK
FROM DUAL
BLOCK
------------
851
System altered.

344
Dump the UNDO Block (Partial)
********************************************************************************
UNDO BLK:
xid: 0x0004.02a.000012ff seq: 0xa9e cnt: 0x7 irb: 0x7 icl: 0x0 flg: 0x0000
...
Rec #0x7 slt: 0x2a objn: 53366(0x0000d076) objd: 53366 tblspc: …
uba: 0x00800353.0a9e.04 ctl max scn: 0x0000.00432655 prv tx scn: 0x0000.00432656
txn start scn: scn: 0x0000.00432731 logon user: 0
prev brb: 8389454 prev bcl: 0
KDO undo record:
KTB Redo
op: 0x03 ver: 0x01
op: Z
KDO Op code: URP row dependencies Disabled
xtype: XA flags: 0x00000000 bdba: 0x0040e1d2 hdba: 0x0040e1d1
itli: 2 ispac: 0 maxfr: 4863
tabn: 0 slot: 1(0x1) flag: 0x2c lock: 0 ckix: 0
ncol: 8 nnew: 1 size: 1
col 1: [ 6] 4d 49 4c 4c 45 52 Here‟s the UNDO: M I L L E R

345
Update „ALLEN‟ to „BIG‟
update emp2
set ename = „BIG‟
where ename = „ALLEN‟;
from emp2;
EMPNO ENAME JOB
-------------- -------------- ---------------------
7499 BIG SALESMAN
7934 SMALL CLERK

346
The ITL again… (fyi to see UBA)
0x01 0xffff.000.00000000 0x00000000.0000.00 C--- 0 scn 0x0000.0043236f
0x02 0x0004.02a.000012ff 0x00800353.0a9e.08 ---- 2 fsc 0x0003.00000000
0x03 0x0000.000.00000000 0x00000000.0000.00 ---- 0 fsc 0x0000.00000000
UBA:
We’re now at Record 8.
Two updates for
the same user
uses the same
ITL. Now save
3 bytes if we
commit.

347
Dump the UNDO Block (Partial)
********************************************************************************
UNDO BLK:
xid: 0x0004.02a.000012ff seq: 0xa9e cnt: 0x7 irb: 0x7 icl: 0x0 flg: 0x0000
...
* Rec #0x7 slt: 0x2a objn: 53366(0x0000d076) objd: 53366 tblspc: 0(0x00000000)
*-----------------------------
uba: 0x00800353.0a9e.04 ctl max scn: 0x0000.00432655 prv tx scn: 0x0000.00432656
txn start scn: scn: 0x0000.00432731 logon user: 0
KDO undo record:
KTB Redo
...
col 1: [ 6] 4d 49 4c 4c 45 52 UNDO RECORD: M I L L E R
* Rec #0x8 slt: 0x2a objn: 53366(0x0000d076) objd: 53366 tblspc: 0(0x00000000)
*-----------------------------
KDO undo record:
KTB Redo
op: C uba: 0x00800353.0a9e.07
...
col 1: [ 5] 41 4c 4c 45 4e UNDO RECORD: A L L E N
End dump data blocks tsn: 1 file#: 2 minblk 851 maxblk 851

348
Now insert some records as user2
insert into emp2
select * from emp1‟;
14 rows created.
System altered.

349
Here is the ITL with our Transaction
0x01 0xffff.000.00000000 0x00000000.0000.00 C--- 0 scn 0x0000.0043236f
0x02 0x0004.02a.000012ff 0x00800353.0a9e.08 ---- 2 fsc 0x0003.00000000
0x03 0x0005.00e.0000029b 0x008005b2.02c9.19 ---- 14 fsc 0x0000.00000000
Transaction ID
Now there are 2 ITL’s in
use.
Undo 4,5 = 4,5 (decimal)
UBA:
2 Undo Headers are used & 2
Undo blocks are referenced.
2 rows were
updated on one
ITL and 14 are
inserted on the
other.

350
Let‟s check V$TRANSACTION &
match it up to ITL (no need to dump)
select xidusn, xidslot, xidsqn, ubafil, ubablk, ubasqn, ubarec
from v$transaction t, v$session s
where t.ses_addr = s.saddr;
XIDUSN XIDSLOT XIDSQN UBAFIL UBABLK UBASQN UBAREC
---------- ---------- ---------- ---------- ---------- ---------- ----------
4 42 4863 2 851 2718 8
5 14 667 2 1458 713 25
4.42.4863 = 4.2a.12ff 2.851.2718.8 = 800353.a9e.8
5.14.667 = 5.e.29b 2.1458.713.25 = 8005b2.2c9.19
0x02 0x0004.02a.000012ff 0x00800353.0a9e.08 ---- 2 fsc 0x0003.00000000
0x03 0x0005.00e.0000029b 0x008005b2.02c9.19 ---- 14 fsc 0x0000.00000000

351
Row Level Locks
(Chain
Latch)
n
1
3
Cache Buffers
Buffer
Header
Hash
Buckets
LRU Lists LRU Write List
DB
Write
1
2
3
Log
Buffers
User
Request
Shared
Pool
SGA
Buffer Cache
2
Cache Layer
2
Transaction Layer
XID 01
XID 02
2 Row 3
1 Row 1
1 Row 2
Lock Byte
ITL 1
ITL 2
XID 01 ITL 1
Itl Xid Uba Flag Lck …
0x01 0x0005.02a.0000028c 0x008000af.026b.01 ---- 1 …
usn#
0x02a 0x028c
RBU Header 5
wrap#
Data Block
slot#
XID
0X005.02a.0000028c
Transaction Identifiers

352
Commit EVERYTHING!
Commit; (all sessions)
System altered.
0x01 0xffff.000.00000000 0x00000000.0000.00 C--- 0 scn 0x0000.0043236f
0x02 0x0004.02a.000012ff 0x00800353.0a9e.08 ---- 2 fsc 0x0003.00000000
0x03 0x0005.00e.0000029a 0x008005b2.02c9.19 ---- 14 fsc 0x0000.00000000
Why no Change (show uncommitted)?? Delayed Block Cleanout! (Usually fast commit)

353
Delayed block cleanout…
Select * from emp2;
(delayed block cleanout is how redo can be generated from a select)
System altered.
0x01 0xffff.000.00000000 0x00000000.0000.00 C--- 0 scn 0x0000.0043236f
0x02 0x0004.02a.000012ff 0x00800353.0a9e.08 C--- 0 scn 0x0000.0043469f
0x03 0x0005.00e.0000029a 0x008005b2.02c9.19 C--- 0 scn 0x0000.004346a3
All records now show as committed.

354
Delayed block cleanout…
If a dirty block has already been written to disk
(could also be due to locking or when many
blocks are changed), then the next process to
visit the block will automatically check the
transaction entry in the undo segment header
and find the changes made to the block have
been commited. The process gets the SCN of
the commit from the undo header transaction
entry and writes it to the data block header to
record the change as commited.

355
What‟s Next?
Indexes (Draft) – FYI quick look...

356
Bitmap Indexes
row#0[8010] flag: ---D-, lock: 2
col 0; len 1; (1): 31
col 1; len 6; (6): 02 40 2d 60 00 00
col 2; len 6; (6): 02 40 2d 60 00 07
col 3; len 1; (1): 00
row#1[7989] flag: ---D-, lock: 2
col 0; len 1; (1): 31
col 1; len 6; (6): 02 40 2d 60 00 00
col 2; len 6; (6): 02 40 2d 60 00 07
col 3; len 2; (2): c8 03
row#2[7968] flag: -----, lock: 2
col 0; len 1; (1): 31
col 1; len 6; (6): 02 40 2d 60 00 00
col 2; len 6; (6): 02 40 2d 60 00 07
col 3; len 2; (2): c8 07
Indexed value consist of 5 lines:
row#0: Row identification
col 0: Indexed value - length in hex.
col 1: Rowid for the first occurrence
of the indexed value.
col 2: Rowid for the last occurrence
of the indexed value
col 3: Actual bitmap which will have
1 where value occurred in range
between first and last rowid,
otherwise 0 in swap byte
notification, first byte usually cx
where x in (8,…,f). When all slots
for cf filled, new segment starts.
Oracle engine will place a lock on
blocks containing indexed value.

357
Insert 64 records / Dump is taken:
row#0[8008] flag: -----, lock: 0
col 0; len 1; (1): 31
col 1; len 6; (6): 02 40 2d 60 00 00
col 2; len 6; (6): 02 40 2d 60 00 3f
col 3; len 9; (9): cf ff ff ff ff ff ff ff ff
Another insert:
row#0[8007] flag: -----, lock: 0
col 0; len 1; (1): 31
col 1; len 6; (6): 02 40 2d 60 00 00
col 2; len 6; (6): 02 40 2d 60 00 40
col 3; len 10; (10): cf ff ff ff ff ff ff ff ff 00
Bitmap Indexes

358
00 will represent first record in a range, i.e.:
col 1; len 6; (6): 02 40 2d 60 00 00
col 2; len 6; (6): 02 40 2d 60 00 40
col 3; len 10; (10): cf ff ff ff ff ff ff ff ff 00
Index value exist in a block which rowid=02 40 2d 60 00 40 (40(hex) =>
64 (decimal)). Previous end rowid was 3f (3*16+15 = 63). When index
is updated there must be enough space to accommodate growth, if
there is no space, split will occur. Not just that, lock will be placed on
entry in leaf index. But this entry can span over multiple blocks. As a
side effect for a duration of a lock no other transaction will be able to
update blocks in affected block range. Bitmap indexes are used for a
column with just a few different values, so on a bigger tables their
bitmap will likely cover quite a few blocks, lock placed on those will
have disastrous effect on other transaction.
Bitmap Indexes

359
Regular index entry was:
col 0; len 2; (2): c1 02
col 1; len 7; (7): 78 69 0c 19 03 27 10
col 2; len 6; (6): 02 40 2e 70 00 00
Reverse key:
col 0; len 2; (2): 02 c1
col 1; len 7; (7): 10 27 03 19 0c 69 78
col 2; len 6; (6): 02 40 2e 70 00 00
Reverse key index will be marked with value
NORMAL/REV in dba_indexes(index_type).
Reverse Key Indexes

360
Trace file on regular index:
col 0; len 1; (1): 61
col 0; len 2; (2): 61 61
col 0; len 3; (3): 61 61 61
col 0; len 4; (4): 61 61 61 61
col 0; len 5; (5): 61 61 61 61 61
col 0; len 6; (6): 61 61 61 61 61 61
col 0; len 7; (7): 61 61 61 61 61 61 61
col 0; len 8; (8): 61 61 61 61 61 61 61 61
col 0; len 9; (9): 61 61 61 61 61 61 61 61 61
col 0; len 10; (10): 61 61 61 61 61 61 61 61 61 61
Trace file on descending index:
col 0; len 10; (10): 9e 9e 9e 9e 9e 9e 9e 9e 9e ff
col 0; len 9; (9): 9e 9e 9e 9e 9e 9e 9e 9e ff
col 0; len 8; (8): 9e 9e 9e 9e 9e 9e 9e ff
col 0; len 7; (7): 9e 9e 9e 9e 9e 9e ff
col 0; len 6; (6): 9e 9e 9e 9e 9e ff
col 0; len 5; (5): 9e 9e 9e 9e ff
col 0; len 4; (4): 9e 9e 9e ff
col 0; len 3; (3): 9e 9e ff
col 0; len 2; (2): 9e ff
Ascending / Descending Index

361
Grid Control – Find Problems FAST
*** TEST SCENARIO ***
Top Activity Shows Lots of Waiting
Issues with Locking

362
Top Activity says BIG PROBLEMS
Almost 200
users are
Active
We have
some DML
Issues

363
Top SQL #1 Details – Locking Issue
Tax
Package
shows row
lock
enqueue
issue
MANY
users are
dividing the
activity

364
Histogram for Enqueues – Long Waits
LONG
waits for
the TX row
lock

365
Top SQL #2 – Update Statement
The query
in the
causing the
locks

366
Top SQL #3 – Insert Statement
The
application
is causing
BIG
problems
Now there
are over
400 active
users

367
Top SQL #3 – Insert Statement
Enqueue waits related to ITL allocations
The Insert
into one of
the TAX
Tables
ITL issues
Some
minor RAC
gc issues

368
Grid Control – Find Problems FAST
*** TEST SCENARIO ***
Go to other Tools to Verify & Advise:
ADDM
Hang Analysis
ASH

369
Go to ADDM to get Verify & Advise
<5 minutes later>
ADDM
sees the
row
contention
for the
Update

370
Grid Control Lessons Learned
Specific
Update
Statement
for Tuning
Results
Suggests to
gather
statistics

371
ASH Report – Points to same issues!
ASH Report For (9:12 AM) ... <17 minutes later to verify>
(1 Report Target Specified)
DB Name DB Id Instance Inst num Release RAC Host
DBRJN 277251124 ORA10 2 10.2.0.3.0 YES linux2
CPUs SGA Size Buffer Cache Shared Pool ASH Buffer Size
8 6,000M (100%) 1,536M (25.6%) 1,025M (17.1%) 29.0M (0.5%)
Top User Events
Event Event Class % Activity Avg Active Sessions
enq: TX - row lock contention Application 99.33 146.11
Top SQL Statements
SQL ID Planhash % Activity Event % Event
SQL Text
F9s5jwdq1j0hz 3197038047 97.32 enq: TX - row lock contention 97.32
UPDATE ASO_QUOTE_HEADERS_ALL S...

372
Go to Hang Analysis & Verify the Pain!
<stay ahead of the problem>
All these
Red &
Yellow
Colors are
NOT a
good sign!

373
Pent-up Demand will crush the system

374
After SOME table fixes & other fixes...
Pent-up Demand causes instant SPIKE!
Within a
single
minute we
have 140
users
waiting
After 10
minutes
there are
240 users
waiting

375
Queueing Pattern – No problem
Spikes up
and then
back down
regularly

376
Queue / De-Queue Process
Spikes up
and then
back down
regularly
Can watch
performance

377
RMAN runs then finishes...
Note the
RMAN
sessions
running
Eventually
ended

378
Temp Space Issue averted ...
Critical
Temp Space
Issue
Added
space

379
Table Growth Issue... HW Enqueue
Insert
causing
issues with
table
growth
Issue is
High Water
Mark on the
Table

380
Temp Space Issue – TS Enqueue
3rd worst
query is a
Temp Space
(TS) issue

381
TCP Socket (KGAS)
Time to
search for
another job

382
Tuning Multiple Nodes
Are all the Nodes up in the Cluster?
5 Nodes All Up! 2 Nodes Down 1 Starting up!

383
Tuning the RAC Cluster Interconnect
– Guidelines for GES Statistics:
• All times should be < 15ms
• Ratio of global lock gets vs global lock releases should be
near 1.0
– High values could indicate possible network or
memory problems
– Could also be caused by application locking issues
– May need to review the enqueue section of
STATSPACK report for further analysis.
Complete Presentation by Oracle’s Rich Niemiec’s at:
http://www.oracleracsig.org

384
Using AWR Reports (FYI Only)

385
Using AWR Reports (FYI Only)

386
High Availability
Handle Issues that Bring Nodes Down

387
Grid Control – One Node goes Down
Issues arise
that must be
fixed
Memory
issues cause
a node to
go down

388
Grid Control – 2nd Node goes Down
Issues
continue
and get
worse
A 2nd node
goes down

389
Grid Control – 3rd Node goes Down
Issues
continue to
get worse
A 3rd node
goes down

390
Grid Control – Rolling Availability
Memory
Issue is
resolved
First node
is back up
and
carrying the
load

391
Node being Evicted NOT
Lack of CPU... Interconnect Failure
Log Files:
$CRS_HOME/log/inst1/cssd/ocssd.log - node 1
$CRS_HOME/log/inst2/cssd/ocssd.log - node 2
If you look at the log file on node 1, which was the master node at that time, you will see :
[ CSSD]2007... >WARNING: clssnmPollingThread: node inst2 (2) at 50% heartbeat fatal, eviction in 29.730 seconds
[ CSSD]2007... >WARNING: clssnmPollingThread: node inst2 (2) at 75% heartbeat fatal, eviction in 14.710 seconds
[ CSSD]2007...>WARNING: clssnmPollingThread: node inst2 (2) at 90% heartbeat fatal, eviction in 5.690 seconds
[ CSSD]2007...>TRACE: clssnmPollingThread: node inst2 (2) is impending reconfig
[ CSSD]2007...>TRACE: clssnmPollingThread: diskTimeout set to (57000)ms impending reconfig status(1)
...etc
[ CSSD]2007...>TRACE: clssnmReadDskHeartbeat: node(2) is down. rcfg(17) ...Disk lastSeqNo(1038569)
[ CSSD]2007...>TRACE: clssnmCheckDskInfo: node(2) disk HB found, network state 0, disk state(3) misstime(770)
[ CSSD]2007...>TRACE: clssnmReadDskHeartbeat: node(2) is down. rcfg(17) ... Disk lastSeqNo(1038570)
[ CSSD]2007...>TRACE: clssnmCheckDskInfo: node(2) misstime(760) state(0). Smaller(1) cluster node 2. mine is 3. (2/1)
[ CSSD]2007...>TRACE: clssnmEvict: Start
[ CSSD]2007...>TRACE: clssnmEvict: Evicting node 2, birth 10, death 17, killme 1
[ CSSD]2007...>TRACE: clssnmEvict: Evicting Node(2), timeout(190)

392
Grid Control – Availability
Tracking
how the
system is
doing...

393
Grid Control – Adding Comments
When you are taking outage you can add comments / reasons.
When you get an alert, go to Grid Control page for target &
edit/update/acknowledge the alerts and add comments to it .
Add a
comment...

394
If Time Permits… the Future!

395
64-Bit advancement of
Directly addressable memory
Address Direct Indirect/Extended
• 4 Bit: 16 (640)
• 8 Bit: 256 (65,536)
• 16 Bit: 65,536 (1,048,576)
• 32 Bit: 4,294,967,296
• 64 Bit: 18,446,744,073,709,551,616
• When the hardware physically implements the theoretical
possibilities of 64-Bit, things will dramatically change….
…moving from 32 bit to 64 bit will be like moving from 4 bit
to 32 bit or like moving from 1971 to 2000 overnight.

You could stack documents from the
Earth so high they would pass Pluto!
Stack single
sheets (2K worth
of text on each)
about 4.8B miles
high to get 16E!!
64bit allows Directly Addressing
16 Exabytes of Memory

397
2K – A typewritten page
5M – The complete works of Shakespeare
10M – One minute of high fidelity sound
2T – Information generated on YouTube in one day
10T – 530,000,000 miles of bookshelves at the Library of Congress
20P – All hard-disk drives in 1995 (or your database in 2010)
700P –Data of 700,000 companies with Revenues less than $200M
1E – Combined Fortune 1000 company databases (average 1P each)
1E –Next 9000 world company databases (average 100T each)
8E – Capacity of ONE Oracle11g Database (CURRENT)
12E to 16E – Info generated before 1999 (memory resident in 64-bit)
16E – Addressable memory with 64-bit (CURRENT)
161E – New information in 2006 (mostly images not stored in DB)
1Z – 1000E (Zettabyte - Grains of sand on beaches -125 Oracle DBs)
100TY - 100T-Yottabytes – Addressable memory 128-bit (FUTURE)
The Future: 8 Exabytes
Look what fits in one 11g Database!

398
What we covered:
• Kramer‟s missing Backup
• George‟s Untuned System
• Elaine‟s Untouched System
• Jerry‟s Perfect Tuning Plan
• Statspack / AWR
– Top Waits
– Load Profile
– Latch Waits
– Top SQL
– Instance Activity
– File I/O
• The Future: EM & ADDM
• Helpful V$/X$ /(Mutexes)
• Exa-Products & Summary

“Perfection is achieved, not when there is nothing left to
add, but when there is nothing left to take away.”
--Antoine de Saint Exupery

Compelling Technology Statistics!
0
5
10
15
20
25
30
35
40
Radio TV Cable Internet Wireless
Years to Reach 50M
Users

402
Friedman’s 6 Dimensions
of Understanding Globalization*
• Politics (Merging)
• Culture (Still disparate)
• Technology (Merging/Merged)
• Finance (Merging/Merged)
• National security (Disparate)
• Ecology (Merging)
* Sited from Mark Hasson, PSU, Global Pricing and International
Marketing.

403
Rich Niemiec, Rolta TUSC (www.roltatusc.com.au)
Oracle 声明: 以下介绍的Oracle产品方向，仅作参考，不具备法律效力，不承诺提
供任何材料、代码或功能，不应作为采购决策依据。Oracle有权自行决定产品
开发、发布和任何功能特性的时间表。
11g最佳新特性
Open World 2007 (基于11g Beta)

406
Before they were DBAs, they
were Engineers

407
This is How DBA View Themselves!

408
Be a Better DBA to the Business!
Do Developers think of this when they think of their
Data

409
Statspack – 仍然有用
一些10g 新特性
开发者想到DBA时，是否就像这样?
Data

411
V$ Views over the years
Version V$ Views X$ Tables
6 23 ? (35)
7 72 126
8.0 132 200
8.1 185 271
9.0 227 352
9.2 259 394
10.1.0.2 340 (+31%) 543 (+38%)
10.2.0.1 396 613
11.1.0.6.0 484 (+23%) 798 (+30%)
11.2.0.1.0 496 (+25%) 945 (+54%)

412
MARK YOUR CALENDARS!
COLLABORATE 12
April 22-26, 2012
Las Vegas, Nevada
http://events.ioug.org/p/cm/ld/fid=15

413
• www.tusc.com
• Oracle 10g Performance
Tuning Tips & Techniques;
Richard J. Niemiec; Oracle
Press (June 2007)
• Oracle 11g Release 2
Performance Tuning Tips &
Techniques (Early 2012)
“If you are going through hell, keep going” - Churchill
For More Information

414
更多信息
• www.tusc.com
• Oracle9i Performance Tuning
Tips & Techniques; Richard
J. Niemiec; Oracle Press (May
2003)
• Oracle 10g Tuning (June 11,
2007)
“成功只访问那些没空追求它的人。”
- Henry David Thoreau

415
References
• www.tusc.com, www.oracle.com, www.ixora.com,
www.laoug.org, www.ioug.org, technet.oracle.com
• Oracle 9i, 10g, 11g documentation
• Oracle10g Performance Tuning Tips & Techniques, Rich Niemiec
• Oracle PL/SQL Tips and Techniques, Joseph P. Trezzo; Oracle Press
• Oracle9i Web Development, Bradley D. Brown; Oracle Press
• Special thanks to Steve Adams, Mike Ault, Brad Brown, Kevin
Gilpin, Herve Lejeune, Randy Swanson and Joe Trezzo.
• Dedicated to the memory of Stan Yellott, Mark Beaton, Ray
Mansfield, Lex De Haan, Elaine DeMeo and Jim Gray.

416
References
• The Self-managing Database: Automatic Performance Diagnosis;
Karl Dias & Mark Ramacher, Oracle Corporation
• EM Grid Control 10g; otn.oracle.com, Oracle Corporation
• Oracle Database 10g Automated Features , Mike Ault, TUSC
• Oracle Enterprise Manager 10g: Making the Grid a Reality; Jay
Rossiter, Oracle Corporation
• The Self-Managing Database: Guided Application and SQL Tuning;
Benoit Dageville, Oracle Corporation
• The New Enterprise Manager: End to End Performance Management
of Oracle; Julie Wong & Arsalan Farooq, Oracle Corporation
• Enterprise Manager : Scalable Oracle Management; John Kennedy,
Oracle Corporation

417
References
• www.tusc.com. www.rolta.com
• Oracle10g Performance Tuning Tips & Techniques; Richard J. Niemiec; Oracle Press
• Database Secure Configuration Initiative: Enhancements with Oracle
Database 11g, www.oracle.com
• All Oracle11g Documentation from Oracle Beta Site
• Introduction to Oracle Database 11g, Ken Jacobs
• Oracle Database 11g New Features, Linda Smith
• New Optimizer Features in 11g, Maria Colgan
• www.ioug.org, www.oracle.com, en.wikipedia.org & technet.oracle.com
• Thanks Dan M., Bob T., Brad, Joe, Heidi, Mike K., Debbie, Maria, Linda
• All companies and product names are trademarks or registered
trademarks of the respective owners.
• Dedicated to the memory of Robert Delgado Patton, Stan Yellott, Mark
Beaton, Ray Mansfield, Lex De Haan, Elaine DeMeo and Jim Gray.

418
V$ View Poster – At the Rolta Booth!

419
Rolta TUSC – Your Partner ….
Accomplished in Oracle!
2011 Oracle Partner of the Year (8 Titans Total)
Prior Years Winner 2002, 2004*, 2007*, 2008, 2011
*Won 2 Awards

420
Rolta TUSC Services
• Oracle
– E-Business Suite implementation, R12 upgrades, migration & support
– Fusion Middleware and Open Systems development
– Business Intelligence (OBIEE) development
– Hyperion Financial Performance Management
– DBA and Database tactical services
– Strategic Global Sourcing
• IT Infrastructure
– IT Roadmap - Security & Compliance - Infrastructure Management
– Enterprise Integration / SOA - High Availability and Disaster Planning
• Profitability & Cost Management
– Financial Consolidation - Budgeting & Forecasting
– Profitability & Risk Analysis - Enterprise Performance Management
– Operational, Financial & Management Reporting
• Rolta Software Solutions
– iPerspective™ - rapid data & systems integration
– Geospatial Fusion™ - spatial integration & visualization
– OneView™ - business & operational intelligence

421
Copyright Information
• Neither Rolta TUSC nor the author guarantee this document
to be error-free. Please provide comments/questions to
rich@tusc.com. I am always looking to improve!
• Rich Niemiec/ Rolta TUSC © 2011. This document cannot
be reproduced without expressed written consent from Rich
Niemiec or an officer of Rolta TUSC, but may be reproduced
or copied for presentation/conference use.
Contact Information
Rich Niemiec: rich@tusc.com
www.tusc.com

422
Rich‟s Overview
(rich@tusc.com)
• Advisor to Rolta International Board
• Former President of TUSC
– Inc. 500 Company (Fastest Growing 500 Private Companies)
– 10 Offices in the United States (U.S.); Based in Chicago
– Oracle Advantage Partner in Tech & Applications
• Former President Rolta TUSC & President Rolta EICT International
• Author (3 Oracle Best Sellers – #1 Oracle Tuning Book for a Decade):
– Oracle Performing Tips & Techniques (Covers Oracle7 & 8i)
– Oracle9i Performance Tips & Techniques
– Oracle Database 10g Performance Tips & Techniques
• Former President of the International Oracle Users Group
• Current President of the Midwest Oracle Users Group
• Chicago Entrepreneur Hall of Fame - 1998
• E&Y Entrepreneur of the Year & National Hall of Fame - 2001
• IOUG Top Speaker in 1991, 1994, 1997, 2001, 2006, 2007
• MOUG Top Speaker Twelve Times
• National Trio Achiever award - 2006
• Oracle Certified Master & Oracle Ace Director
• Purdue Outstanding Electrical & Computer and Engineer - 2007

ohio11_grid_kramer_present.pdf Statspack

More Related Content

Similar to ohio11_grid_kramer_present.pdf Statspack (20)

More from cookie1969 (20)

Recently uploaded (20)

ohio11_grid_kramer_present.pdf Statspack