Operating system virtual memory by silberscatz

Silberschatz, Galvin and Gagne ©2013
Operating System Concepts – 9th
Edition
Chapter 9: Virtual Memory

9.2 Silberschatz, Galvin and Gagne ©2013
Edition
Chapter 9: Virtual Memory
 Background
 Demand Paging
 Copy-on-Write
 Page Replacement
 Allocation of Frames
 Thrashing
 Memory-Mapped Files
 Allocating Kernel Memory
 Other Considerations
 Operating-System Examples

Edition
Objectives
 To describe the benefits of a virtual memory system
 To explain the concepts of demand paging, page-replacement
algorithms, and allocation of page frames
 To discuss the principle of the working-set model
 To examine the relationship between shared memory and
memory-mapped files
 To explore how kernel memory is managed

Edition
Background
 Code needs to be in memory to execute, but entire program rarely
used
 Error code, unusual routines, large data structures
 Entire program code not needed at same time
 Consider ability to execute partially-loaded program
 Program no longer constrained by limits of physical memory
 Each program takes less memory while running -> more
programs run at the same time
 Increased CPU utilization and throughput with no increase
in response time or turnaround time
 Less I/O needed to load or swap programs into memory ->
each user program runs faster

Edition
Background (Cont.)
 Virtual memory – separation of user logical memory from
physical memory
 Only part of the program needs to be in memory for execution
 Logical address space can therefore be much larger than physical
address space
 Allows address spaces to be shared by several processes
 Allows for more efficient process creation
 More programs running concurrently
 Less I/O needed to load or swap processes

Edition
Background (Cont.)
 Virtual address space – logical view of how process is stored
in memory
 Usually start at address 0, contiguous addresses until end of
space
 Meanwhile, physical memory organized in page frames
 MMU must map logical to physical
 Virtual memory can be implemented via:
 Demand paging
 Demand segmentation

Edition
Virtual Memory That is Larger Than Physical Memory

Edition
Virtual-address Space
 Usually design logical address space for
stack to start at Max logical address and
grow “down” while heap grows “up”
 Maximizes address space use
 Unused address space between
the two is hole
 No physical memory needed
until heap or stack grows to a
given new page
 Enables sparse address spaces with
holes left for growth, dynamically linked
libraries, etc
 System libraries shared via mapping into
virtual address space
 Shared memory by mapping pages read-
write into virtual address space
 Pages can be shared during fork(),
speeding process creation

Edition
Shared Library Using Virtual Memory

Edition
Demand Paging
 Could bring entire process into memory
at load time
 Or bring a page into memory only when
it is needed
 Less I/O needed, no unnecessary
I/O
 Less memory needed
 Faster response
 More users
 Similar to paging system with swapping
(diagram on right)
 Page is needed  reference to it
 invalid reference  abort
 not-in-memory  bring to memory
 Lazy swapper – never swaps a page
into memory unless page will be needed
 Swapper that deals with pages is a
pager

Edition
Basic Concepts
 With swapping, pager guesses which pages will be used before
swapping out again
 Instead, pager brings in only those pages into memory
 How to determine that set of pages?
 Need new MMU functionality to implement demand paging
 If pages needed are already memory resident
 No difference from non demand-paging
 If page needed and not memory resident
 Need to detect and load the page into memory from storage
 Without changing program behavior
 Without programmer needing to change code

Edition
Valid-Invalid Bit
 With each page table entry a valid–invalid bit is associated
(v  in-memory – memory resident, i  not-in-memory)
 Initially valid–invalid bit is set to i on all entries
 Example of a page table snapshot:
 During MMU address translation, if valid–invalid bit in page table
entry is i  page fault

Edition
Page Table When Some Pages Are Not in Main Memory

Edition
Page Fault
 If there is a reference to a page, first reference to that page will
trap to operating system:
page fault
1. Operating system looks at another table to decide:
 Invalid reference  abort
 Just not in memory
2. Find free frame
3. Swap page into frame via scheduled disk operation
4. Reset tables to indicate page now in memory
Set validation bit = v
5. Restart the instruction that caused the page fault

Edition
Steps in Handling a Page Fault

Edition
Aspects of Demand Paging
 Extreme case – start process with no pages in memory
 OS sets instruction pointer to first instruction of process, non-
memory-resident -> page fault
 And for every other process pages on first access
 Pure demand paging
 Actually, a given instruction could access multiple pages -> multiple
page faults
 Consider fetch and decode of instruction which adds 2 numbers
from memory and stores result back to memory
 Pain decreased because of locality of reference
 Hardware support needed for demand paging
 Page table with valid / invalid bit
 Secondary memory (swap device with swap space)
 Instruction restart

Edition
Instruction Restart
 Consider an instruction that could access several different locations
 block move
 auto increment/decrement location
 Restart the whole operation?
 What if source and destination overlap?

Edition
Performance of Demand Paging
 Stages in Demand Paging (worse case)
1. Trap to the operating system
2. Save the user registers and process state
3. Determine that the interrupt was a page fault
4. Check that the page reference was legal and determine the location of the page on the disk
5. Issue a read from the disk to a free frame:
1. Wait in a queue for this device until the read request is serviced
2. Wait for the device seek and/or latency time
3. Begin the transfer of the page to a free frame
6. While waiting, allocate the CPU to some other user
7. Receive an interrupt from the disk I/O subsystem (I/O completed)
8. Save the registers and process state for the other user
9. Determine that the interrupt was from the disk
10. Correct the page table and other tables to show page is now in memory
11. Wait for the CPU to be allocated to this process again
12. Restore the user registers, process state, and new page table, and then resume the interrupted
instruction

Edition
Performance of Demand Paging (Cont.)
 Three major activities
 Service the interrupt – careful coding means just several hundred instructions
needed
 Read the page – lots of time
 Restart the process – again just a small amount of time
 Page Fault Rate 0  p  1
 if p = 0 no page faults
 if p = 1, every reference is a fault
 Effective Access Time (EAT)
EAT = (1 – p) x memory access
+ p (page fault overhead
+ swap page out
+ swap page in )

Edition
Demand Paging Example
 Memory access time = 200 nanoseconds
 Average page-fault service time = 8 milliseconds
 EAT = (1 – p) x 200 + p (8 milliseconds)
= (1 – p x 200 + p x 8,000,000
= 200 + p x 7,999,800
 If one access out of 1,000 causes a page fault, then
EAT = 8.2 microseconds.
This is a slowdown by a factor of 40!!
 If want performance degradation < 10 percent
 220 > 200 + 7,999,800 x p
20 > 7,999,800 x p
 p < .0000025
 < one page fault in every 400,000 memory accesses

Edition
Demand Paging Optimizations
 Swap space I/O faster than file system I/O even if on the same device
 Swap allocated in larger chunks, less management needed than file system
 Copy entire process image to swap space at process load time
 Then page in and out of swap space
 Used in older BSD Unix
 Demand page in from program binary on disk, but discard rather than paging out
when freeing frame
 Used in Solaris and current BSD
 Still need to write to swap space
 Pages not associated with a file (like stack and heap) – anonymous
memory
 Pages modified in memory but not yet written back to the file system
 Mobile systems
 Typically don’t support swapping
 Instead, demand page from file system and reclaim read-only pages (such
as code)

Edition
Copy-on-Write
 Copy-on-Write (COW) allows both parent and child processes to initially
share the same pages in memory
 If either process modifies a shared page, only then is the page copied
 COW allows more efficient process creation as only modified pages are
copied
 In general, free pages are allocated from a pool of zero-fill-on-demand
pages
 Pool should always have free frames for fast demand page execution
 Don’t want to have to free a frame as well as other processing on
page fault
 Why zero-out a page before allocating it?
 vfork() variation on fork() system call has parent suspend and child
using copy-on-write address space of parent
 Designed to have child call exec()
 Very efficient

Edition
Before Process 1 Modifies Page C

Edition
After Process 1 Modifies Page C

Edition
What Happens if There is no Free Frame?
 Used up by process pages
 Also in demand from the kernel, I/O buffers, etc
 How much to allocate to each?
 Page replacement – find some page in memory, but not really in
use, page it out
 Algorithm – terminate? swap out? replace the page?
 Performance – want an algorithm which will result in minimum
number of page faults
 Same page may be brought into memory several times

Edition
Page Replacement
 Prevent over-allocation of memory by modifying page-
fault service routine to include page replacement
 Use modify (dirty) bit to reduce overhead of page
transfers – only modified pages are written to disk
 Page replacement completes separation between logical
memory and physical memory – large virtual memory can
be provided on a smaller physical memory

Edition
Need For Page Replacement

Edition
Basic Page Replacement
1. Find the location of the desired page on disk
2. Find a free frame:
- If there is a free frame, use it
- If there is no free frame, use a page replacement algorithm to
select a victim frame
- Write victim frame to disk if dirty
3. Bring the desired page into the (newly) free frame; update the page
and frame tables
4. Continue the process by restarting the instruction that caused the trap
Note now potentially 2 page transfers for page fault – increasing EAT

Edition
Page Replacement

Edition
Page and Frame Replacement Algorithms
 Frame-allocation algorithm determines
 How many frames to give each process
 Which frames to replace
 Page-replacement algorithm
 Want lowest page-fault rate on both first access and re-access
 Evaluate algorithm by running it on a particular string of memory
references (reference string) and computing the number of page
faults on that string
 String is just page numbers, not full addresses
 Repeated access to the same page does not cause a page fault
 Results depend on number of frames available
 In all our examples, the reference string of referenced page
numbers is
7,0,1,2,0,3,0,4,2,3,0,3,0,3,2,1,2,0,1,7,0,1

Edition
Graph of Page Faults Versus The Number of Frames

Edition
First-In-First-Out (FIFO) Algorithm
 Reference string: 7,0,1,2,0,3,0,4,2,3,0,3,0,3,2,1,2,0,1,7,0,1
 3 frames (3 pages can be in memory at a time per process)
 Can vary by reference string: consider 1,2,3,4,1,2,5,1,2,3,4,5
 Adding more frames can cause more page faults!
 Belady’s Anomaly
 How to track ages of pages?
 Just use a FIFO queue
15 page faults

Edition
FIFO Illustrating Belady’s Anomaly

Edition
Optimal Algorithm
 Replace page that will not be used for longest period of time
 9 is optimal for the example
 How do you know this?
 Can’t read the future
 Used for measuring how well your algorithm performs

Edition
Least Recently Used (LRU) Algorithm
 Use past knowledge rather than future
 Replace page that has not been used in the most amount of time
 Associate time of last use with each page
 12 faults – better than FIFO but worse than OPT
 Generally good algorithm and frequently used
 But how to implement?

Edition
LRU Algorithm (Cont.)
 Counter implementation
 Every page entry has a counter; every time page is referenced
through this entry, copy the clock into the counter
 When a page needs to be changed, look at the counters to find
smallest value
 Search through table needed
 Stack implementation
 Keep a stack of page numbers in a double link form:
 Page referenced:
 move it to the top
 requires 6 pointers to be changed
 But each update more expensive
 No search for replacement
 LRU and OPT are cases of stack algorithms that don’t have
Belady’s Anomaly

Edition
Use Of A Stack to Record Most Recent Page References

Edition
LRU Approximation Algorithms
 LRU needs special hardware and still slow
 Reference bit
 With each page associate a bit, initially = 0
 When page is referenced bit set to 1
 Replace any with reference bit = 0 (if one exists)
 We do not know the order, however
 Second-chance algorithm
 Generally FIFO, plus hardware-provided reference bit
 Clock replacement
 If page to be replaced has
 Reference bit = 0 -> replace it
 reference bit = 1 then:
– set reference bit 0, leave page in memory
– replace next page, subject to same rules

Edition
Second-Chance (clock) Page-Replacement Algorithm

Edition
Enhanced Second-Chance Algorithm
 Improve algorithm by using reference bit and modify bit (if
available) in concert
 Take ordered pair (reference, modify)
1. (0, 0) neither recently used not modified – best page to replace
2. (0, 1) not recently used but modified – not quite as good, must
write out before replacement
3. (1, 0) recently used but clean – probably will be used again soon
4. (1, 1) recently used and modified – probably will be used again
soon and need to write out before replacement
 When page replacement called for, use the clock scheme but
use the four classes replace page in lowest non-empty class
 Might need to search circular queue several times

Edition
Counting Algorithms
 Keep a counter of the number of references that have been made
to each page
 Not common
 Lease Frequently Used (LFU) Algorithm: replaces page with
smallest count
 Most Frequently Used (MFU) Algorithm: based on the argument
that the page with the smallest count was probably just brought in
and has yet to be used

Edition
Page-Buffering Algorithms
 Keep a pool of free frames, always
 Then frame available when needed, not found at fault time
 Read page into free frame and select victim to evict and add
to free pool
 When convenient, evict victim
 Possibly, keep list of modified pages
 When backing store otherwise idle, write pages there and set
to non-dirty
 Possibly, keep free frame contents intact and note what is in them
 If referenced again before reused, no need to load contents
again from disk
 Generally useful to reduce penalty if wrong victim frame
selected

Edition
Applications and Page Replacement
 All of these algorithms have OS guessing about future page
access
 Some applications have better knowledge – i.e. databases
 Memory intensive applications can cause double buffering
 OS keeps copy of page in memory as I/O buffer
 Application keeps page in memory for its own work
 Operating system can given direct access to the disk, getting out
of the way of the applications
 Raw disk mode
 Bypasses buffering, locking, etc

Edition
Allocation of Frames
 Each process needs minimum number of frames
 Example: IBM 370 – 6 pages to handle SS MOVE instruction:
 instruction is 6 bytes, might span 2 pages
 2 pages to handle from
 2 pages to handle to
 Maximum of course is total frames in the system
 Two major allocation schemes
 fixed allocation
 priority allocation
 Many variations

Edition
Fixed Allocation
 Equal allocation – For example, if there are 100 frames (after
allocating frames for the OS) and 5 processes, give each process
20 frames
 Keep some as free frame buffer pool
 Proportional allocation – Allocate according to the size of process
 Dynamic as degree of multiprogramming, process sizes
change
m
S
s
p
a
m
s
S
p
s
i
i
i
i
i
i







for
allocation
frames
of
number
total
process
of
size

Edition
Priority Allocation
 Use a proportional allocation scheme using priorities rather
than size
 If process Pi generates a page fault,
 select for replacement one of its frames
 select for replacement a frame from a process with lower
priority number

Edition
Global vs. Local Allocation
 Global replacement – process selects a replacement frame
from the set of all frames; one process can take a frame from
another
 But then process execution time can vary greatly
 But greater throughput so more common
 Local replacement – each process selects from only its own
set of allocated frames
 More consistent per-process performance
 But possibly underutilized memory

Edition
Non-Uniform Memory Access
 So far all memory accessed equally
 Many systems are NUMA – speed of access to memory varies
 Consider system boards containing CPUs and memory,
interconnected over a system bus
 Optimal performance comes from allocating memory “close to”
the CPU on which the thread is scheduled
 And modifying the scheduler to schedule the thread on the
same system board when possible
 Solved by Solaris by creating lgroups
 Structure to track CPU / Memory low latency groups
 Used my schedule and pager
 When possible schedule all threads of a process and
allocate all memory for that process within the lgroup

Edition
Thrashing
 If a process does not have “enough” pages, the page-fault rate is
very high
 Page fault to get page
 Replace existing frame
 But quickly need replaced frame back
 This leads to:
 Low CPU utilization
 Operating system thinking that it needs to increase the
degree of multiprogramming
 Another process added to the system
 Thrashing  a process is busy swapping pages in and out

Edition
Thrashing (Cont.)

Edition
Demand Paging and Thrashing
 Why does demand paging work?
Locality model
 Process migrates from one locality to another
 Localities may overlap
 Why does thrashing occur?
 size of locality > total memory size
 Limit effects by using local or priority page replacement

Edition
Locality In A Memory-Reference Pattern

Edition
Working-Set Model
   working-set window  a fixed number of page references
Example: 10,000 instructions
 WSSi (working set of Process Pi) =
total number of pages referenced in the most recent  (varies in time)
 if  too small will not encompass entire locality
 if  too large will encompass several localities
 if  =   will encompass entire program
 D =  WSSi  total demand frames
 Approximation of locality
 if D > m  Thrashing
 Policy if D > m, then suspend or swap out one of the processes

Edition
Keeping Track of the Working Set
 Approximate with interval timer + a reference bit
 Example:  = 10,000
 Timer interrupts after every 5000 time units
 Keep in memory 2 bits for each page
 Whenever a timer interrupts copy and sets the values of all
reference bits to 0
 If one of the bits in memory = 1  page in working set
 Why is this not completely accurate?
 Improvement = 10 bits and interrupt every 1000 time units

Edition
Page-Fault Frequency
 More direct approach than WSS
 Establish “acceptable” page-fault frequency (PFF) rate and
use local replacement policy
 If actual rate too low, process loses frame
 If actual rate too high, process gains frame

Edition
Working Sets and Page Fault Rates
 Direct relationship between working set of a process and its
page-fault rate
 Working set changes over time
 Peaks and valleys over time

Edition
Memory-Mapped Files
 Memory-mapped file I/O allows file I/O to be treated as routine
memory access by mapping a disk block to a page in memory
 A file is initially read using demand paging
 A page-sized portion of the file is read from the file system into
a physical page
 Subsequent reads/writes to/from the file are treated as ordinary
memory accesses
 Simplifies and speeds file access by driving file I/O through memory
rather than read() and write() system calls
 Also allows several processes to map the same file allowing the
pages in memory to be shared
 But when does written data make it to disk?
 Periodically and / or at file close() time
 For example, when the pager scans for dirty pages

Edition
Memory-Mapped File Technique for all I/O
 Some OSes uses memory mapped files for standard I/O
 Process can explicitly request memory mapping a file via mmap()
system call
 Now file mapped into process address space
 For standard I/O (open(), read(), write(), close()), mmap
anyway
 But map file into kernel address space
 Process still does read() and write()
 Copies data to and from kernel space and user space
 Uses efficient memory management subsystem
 Avoids needing separate subsystem
 COW can be used for read/write non-shared pages
 Memory mapped files can be used for shared memory (although again
via separate system calls)

Edition
Memory Mapped Files

Edition
Shared Memory via Memory-Mapped I/O

Edition
Shared Memory in Windows API
 First create a file mapping for file to be mapped
 Then establish a view of the mapped file in process’s virtual
address space
 Consider producer / consumer
 Producer create shared-memory object using memory mapping
features
 Open file via CreateFile(), returning a HANDLE
 Create mapping via CreateFileMapping() creating a
named shared-memory object
 Create view via MapViewOfFile()
 Sample code in Textbook

Edition
Allocating Kernel Memory
 Treated differently from user memory
 Often allocated from a free-memory pool
 Kernel requests memory for structures of varying sizes
 Some kernel memory needs to be contiguous
 I.e. for device I/O

Edition
Buddy System
 Allocates memory from fixed-size segment consisting of physically-
contiguous pages
 Memory allocated using power-of-2 allocator
 Satisfies requests in units sized as power of 2
 Request rounded up to next highest power of 2
 When smaller allocation needed than is available, current chunk split
into two buddies of next-lower power of 2
 Continue until appropriate sized chunk available
 For example, assume 256KB chunk available, kernel requests 21KB
 Split into AL and AR of 128KB each
 One further divided into BL and BR of 64KB
– One further into CL and CR of 32KB each – one used to satisfy
request
 Advantage – quickly coalesce unused chunks into larger chunk
 Disadvantage - fragmentation

Edition
Buddy System Allocator

Edition
Slab Allocator
 Alternate strategy
 Slab is one or more physically contiguous pages
 Cache consists of one or more slabs
 Single cache for each unique kernel data structure
 Each cache filled with objects – instantiations of the data
structure
 When cache created, filled with objects marked as free
 When structures stored, objects marked as used
 If slab is full of used objects, next object allocated from empty
slab
 If no empty slabs, new slab allocated
 Benefits include no fragmentation, fast memory request
satisfaction

Edition
Slab Allocation

Edition
Slab Allocator in Linux
 For example process descriptor is of type struct task_struct
 Approx 1.7KB of memory
 New task -> allocate new struct from cache
 Will use existing free struct task_struct
 Slab can be in three possible states
1. Full – all used
2. Empty – all free
3. Partial – mix of free and used
 Upon request, slab allocator
1. Uses free struct in partial slab
2. If none, takes one from empty slab
3. If no empty slab, create new empty

Edition
Slab Allocator in Linux (Cont.)
 Slab started in Solaris, now wide-spread for both kernel mode and
user memory in various OSes
 Linux 2.2 had SLAB, now has both SLOB and SLUB allocators
 SLOB for systems with limited memory
 Simple List of Blocks – maintains 3 list objects for small,
medium, large objects
 SLUB is performance-optimized SLAB removes per-CPU
queues, metadata stored in page structure

Edition
Other Considerations -- Prepaging
 Prepaging
 To reduce the large number of page faults that occurs at
process startup
 Prepage all or some of the pages a process will need, before
they are referenced
 But if prepaged pages are unused, I/O and memory was wasted
 Assume s pages are prepaged and α of the pages is used
 Is cost of s * α save pages faults > or < than the cost of
prepaging
s * (1- α) unnecessary pages?
 α near zero  prepaging loses

Edition
Other Issues – Page Size
 Sometimes OS designers have a choice
 Especially if running on custom-built CPU
 Page size selection must take into consideration:
 Fragmentation
 Page table size
 Resolution
 I/O overhead
 Number of page faults
 Locality
 TLB size and effectiveness
 Always power of 2, usually in the range 212
(4,096 bytes) to 222
(4,194,304 bytes)
 On average, growing over time

Edition
Other Issues – TLB Reach
 TLB Reach - The amount of memory accessible from the TLB
 TLB Reach = (TLB Size) X (Page Size)
 Ideally, the working set of each process is stored in the TLB
 Otherwise there is a high degree of page faults
 Increase the Page Size
 This may lead to an increase in fragmentation as not all
applications require a large page size
 Provide Multiple Page Sizes
 This allows applications that require larger page sizes the
opportunity to use them without an increase in fragmentation

Edition
Other Issues – Program Structure
 Program structure
 int[128,128] data;
 Each row is stored in one page
 Program 1
for (j = 0; j <128; j++)
for (i = 0; i < 128; i++)
data[i,j] = 0;
128 x 128 = 16,384 page faults
 Program 2
for (i = 0; i < 128; i++)
for (j = 0; j < 128; j++)
data[i,j] = 0;
128 page faults

Edition
Other Issues – I/O interlock
 I/O Interlock – Pages must
sometimes be locked into memory
 Consider I/O - Pages that are used
for copying a file from a device
must be locked from being selected
for eviction by a page replacement
algorithm
 Pinning of pages to lock into
memory

Edition
Operating System Examples
 Windows
 Solaris

Edition
Windows
 Uses demand paging with clustering. Clustering brings in pages
surrounding the faulting page
 Processes are assigned working set minimum and working set
maximum
 Working set minimum is the minimum number of pages the
process is guaranteed to have in memory
 A process may be assigned as many pages up to its working set
maximum
 When the amount of free memory in the system falls below a
threshold, automatic working set trimming is performed to
restore the amount of free memory
 Working set trimming removes pages from processes that have
pages in excess of their working set minimum

Edition
Solaris
 Maintains a list of free pages to assign faulting processes
 Lotsfree – threshold parameter (amount of free memory) to
begin paging
 Desfree – threshold parameter to increasing paging
 Minfree – threshold parameter to being swapping
 Paging is performed by pageout process
 Pageout scans pages using modified clock algorithm
 Scanrate is the rate at which pages are scanned. This ranges
from slowscan to fastscan
 Pageout is called more frequently depending upon the amount of
free memory available
 Priority paging gives priority to process code pages

Edition
Solaris 2 Page Scanner

Operating system virtual memory by silberscatz

More Related Content

Similar to Operating system virtual memory by silberscatz (20)

More from rajirajesh8 (6)

Recently uploaded (20)

Operating system virtual memory by silberscatz