Perf by design

 Rico Mariani
Architect
Microsoft Corporation





















 http://blogs.msdn.com/ricom

 http://msdn.microsoft.com/en-us/library/aa338212.aspx

 http://blogs.msdn.com/maoni/

 http://blogs.msdn.com/tess/

© 2008 Microsoft Corporation. All rights reserved. Microsoft, Windows, Windows Vista and other product names are or may be registered trademarks and/or trademarks in the U.S. and/or other countries.
The information herein is for informational purposes only and represents the current view of Microsoft Corporation as of the date of this presentation. Because Microsoft must respond to changing market
conditions, it should not be interpreted to be a commitment on the part of Microsoft, and Microsoft cannot guarantee the accuracy of any information provided after the date of this presentation.
MICROSOFT MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN THIS PRESENTATION.

Vance Morrison
Performance Architect




Stopwatch sw = Stopwatch.StartNew();
// Something being measured
sw.Stop();
Console.WriteLine("Time = {0:f3} MSec", sw.Elapsed.TotalMilliseconds);







1) Go to Control panel’s
Power Options 2) Set to High Performance
•If you don’t do this, CPU is typically throttled to save power
•Throttling causes less consistent measurements



























Related Sessions
Session Title Speaker Day Time Location
PC53 Lunch: Building High Performance JScript Apps Sameer
Chabungbam
Mon 12:45-1:30PM 515B
Microsoft Visual Studio: Bringing out the Best in Multicore Systems Hazim Shafi Mon 1:45-3:00PM 502A
WCF: Zen of Performance and Scale Nicholas Allen Tues 12:45-1:30PM 515B
SQL Server 2008: Developing Large Scale Web Applications and Services Jose Blakeley Tues 1:45-3:00PM 411
Using Instrumentation and Diagnostics to Develop High Quality Software Ricky Buch Tues 5:15-6:30PM 408B
TL24 Improving .NET Application Performance and Scalability Steve Carol Wed 1:15-2:30PM 153
Parallel Programming for Managed Developers with the Next Version of
Microsoft Visual Studio
Daniel Moth Wed 10:30-11:45AM Petree
Hall
HOL Code Title
TLHOL11 VSTS 2010: Diagnostics and Performance
Related Labs


L1 Cache L2 Cache Memory Disk
Size 64K 512K 2000M 100,000M
Access Time .4ns 4ns 40ns 4,000,000ns

























Select Columns
Working Set tends to Overestimates Memory Impact (shared OS files)
But does not account for read files. Private Working Set Underestimates Impact
Small < 20Meg WS Med ~ 50 Meg WS Large > 100 Meg WS
Small < 5 Meg Private Med ~ 20 Meg Private Large > 50 Meg Private

1 Open .NET CLR Memory 2 Select Counters 4 Add 5 OK
3 Select Process

Set Display to Report GC Heap Size
So 7.3 Meg of the 8.6 Meg of private working set is the GC Heap
% Time in GC ideally less than 10%
Ratios of GC Generations Gen0 = 10 x Gen 1, Gen 1 = 10 x Gen 2
% Time in GC


1 Total WS
2 Breakdown
3 DLL Breakdown
GC Heap Here
Only These Affect
Cold Startup











 In Either Case when Working Set Large (> 10Meg)
 Throughput is lost due to cache misses
 Server workloads are typically cache limited









 Thus it Pays to Think about Memory Early!















































 Finding Anomalous Memory use (AKA Leaks)
XmlView Demo











6
0

For all the same Reasons it is for People














6
4
int loopCount = 10;
for (int tId = 1; tId <= 3; tId++) { // Create three work items.
ThreadPool.QueueUserWorkItem(delegate
{
for (int i = 0; i < loopCount; i++)
{
Console.WriteLine("Thread i={0} time: {1}", i, DateTime.Now);
Thread.Sleep(1000);
}
});
}
Console.WriteLine("Waiting for workers. Hit return to end program.");
Console.ReadLine();
You can capture variables
Anonymous Delegate

6
6
0 5 10 15 20 25 30 35 40 45 50
0
20
40
60
80
100
120
140
160
Concurrency Level
Throughput
Work items with 10% CPU on a dual core system
using ThreadPool.QueueUserWorkItem().
Reasons for perf degradation
1. Context switches
2. Memory contention
3. Hot Locks
4. Disk Contention
5. Network Contention

Task task1 = Task.Create(delegate
{
Console.WriteLine("In task 1");
Thread.Sleep(1000); // do work
Console.WriteLine("Done task 1");
});
Task task2 = Task.Create(delegate
{
Console.WriteLine("In task 2");
Console.WriteLine("Done task 2");
});
If (userAbort) { task1.Cancel(); task2.Cancel(); }
Task.WaitAll(task1, task2);
Tasks now have handles
Exceptions in tasks propagated on wait
You can cancel work
You can wait on the
handle to synchronize

Future<int> left = Future.Create(delegate
{
return 1;
});
Future<int> right = Future.Create(delegate
{
return 1;
});
Console.WriteLine("Answer {0}", left.Value + right.Value);
Left and right
represent integers,
but they may not be
computed yet
Asking for the value
forces the wait if the
future is not already
done
In this case the right
value was done, so no
wait was needed

Loops






for(int i = 0; i < n; i++)
{
work(i);
}
foreach(T e in data)
{
work(e);
}
Parallel.For(0, n, i =>
{
work(i);
}
Parallel.ForEach(data, e =>
{
work(e);
}

 Enable LINQ developers to leverage parallel
hardware






 Augments LINQ-to-Objects, doesn’t replace it















Related Sessions
Session Title Speaker Day Time Location
Microsoft Visual Studio: Bringing out the Best in Multicore Systems Hazim Shafi Mon 1:45-3:00PM 502A
TL24 Improving .NET Application Performance and Scalability Steve Carol Wed 1:15-2:30PM 153
Parallel Programming for Managed Developers with the Next Version of
Microsoft Visual Studio
Daniel Moth Wed 10:30-11:45AM Petree
Hall
Parallel Symposium: Addressing the Hard Problems with Concurrency David Callahan Thurs 8:30-10:00AM 515A
Parallel Symposium: Future of Parallel Computing Dave Detlefs Thurs 12:00-1:30PM 515A

 Mark Friedman
Architect
Developer Division











78


Dynamic



<A HREF="resumepage.html">my resume</A>





<LINK REL=STYLESHEET HREF="mystyles.css"
TYPE="text/css">
79










connectionless sessionless


80


 Postback
 HttpContext

 ViewState
 Session State


81


 On the client
 Performance counters
 Network Monitor traces
 ETW traces (IE, etc.)
 Direct measurements inside Javascript code
 On the server
 IIS logs (especially Time-Taken fields)
 Performance counters
 web app response time data is not available
 ETW traces (IIS, ASP.NET, CLR, etc.)
 volume considerations
 Direct measurement inside .NET code
 e.g., Stopwatch
83


 End-to-End
 Multiple tiers
84


network latency
Network
Latency
Client-side
script
Server-side
.aspx
Network
Latency
Client-side
script
Server-side
.aspx
Network
Latency
Client-side
script
Server-side
.aspx
Unit Test Load Test
e.g., VS TeamTest
Production
85

 Page Load Time









 Ready 86














91









 Semi-dynamic
92


may be






93










 effective
 Caching Architecture Guide for .NET Framework
Applications
94

w3wp.exe
Common Language Runtime (CLR)
JIT compiler
Garbage Collection threads
mscoree.dll
mscorsvr.dll
MyApp.dll
95

 runat=“server”
Form:
<asp:DropDownList id="FileDropDownList" style="Z-INDEX: 111; LEFT:
192px; POSITION: absolute; TOP: 240px“ runat="server" Width="552px"
Height="40px“ AutoPostBack="True"></asp:DropDownList>
Code behind:
Private Sub FileDropDownList_SelectedIndexChanged _
(ByVal sender As System.Object, _
ByVal e As System.EventArgs) _
Handles FileDropDownList.SelectedIndexChanged
Session("FileSelected") = FileDropDownList.SelectedValue
End Sub
96

97
IIS Architecture
User
Kernel
HTTP
TCP
IP
Network
Interface
HTTP Kernel Mode
Driver
(http.sys)
HTTP Response Cache
(Physical Memory)
LSSAS
Windows
Authentication
SSL
Inetinfo
IIS
Administration
FTP
Metabase
SMTP
NNTP
SVCHOST
W3SVC
W3wp
W3wp
Mscoree.dll
http
<code-behind>.dll
Default.aspx
Application Pool
W3wp
W3wp
Mscoree.dll
<code-behind>.dll
Default.aspx
Application Pool
WAS
Cache
net.tcp net.tcp
http
HTTP Request

98

http.sys

Response object cache


 Web Service
CacheKernel: Current
URIs Cached, etc.
See “IIS 7.0 Output Caching”
at http://learn.iis.net/page.aspx/154/iis-7-output-caching/

100
 w3wp.exe

From a command line:
windowssystem32inetsrvappcmd list WP








Request Life-cycle Events
Begin Request
Authentication
Authorization
Resolve Cache
Map Handler
Acquire State
Execute Handler
Release State
Update Cache
Log Request
Send Request
104


Event Event
BeginRequest PreRequestHandlerExecute
AuthenticateRequest PostRequestHandlerExecute
PostAuthenticateRequest ReleaseRequestState
AuthorizeRequest PostReleaseRequestState
PostAuthorizeRequest PostMapRequestHandler
ResolveRequestCache PostMapRequestHandler
PostResolveRequestCache PostMapRequestHandler
MapRequestHandler UpdateRequestCache
PostMapRequestHandler PostUpdateRequestCache
AcquireRequestState LogRequest
PostAcquireRequestState EndRequest 105






 <modules>




106

RequestNotification
 Mark Friedman
Architect
Developer Division

110

Level = 5 (Diagnostic level)

111

 tracerpt
tracerpt filename.etl –of CSV –o mytrace.csv
 logparser
logparser "select * from filename.etl" -e 20
-o:DATAGRID -rtp 30 -compactModeSep "|"
 xperf

114

 UseUrlFilter to control the volume of data
 Configure the TraceUriPrefix
See “How to Trace Requests for a
Specific URL or Set of URLs” at
http://www.microsoft.com/technet/pro
dtechnol/WindowsServer2003/Library/II
S/c56d19af-b3d1-4be9-8a6f-
4aa86bacac3f.mspx?mfr=true

115




 HttpContext

 Request
 Response
 Session
 Cache

116

 HttpContext.Request
 HttpMethod (GET, POST, etc.)
 URL
 Cookies collection
 Headers
 InputStream
 UserHostAddress (IP address of the Requestor)
 etc.
 The ASP.NET programming model provides several
facilities to persist User/Application state

Event Usage
PreInit Create dynamic controls, set the Theme; master page, etc.
Init Read or initialize control properties
InitComplete Raised by the Page object.
PreLoad Perform any processing on your page or control before the Load event.
Load The Page calls the OnLoad event method on the Page, then recursively for each
child control and its children
Control events Button Clicks and other Control events are processed after Page_Load
LoadComplete Fires after all controls on the page are loaded.
PreRender Data binding for controls occurs now.
SaveStateComplete Fires when the ViewState for all controls is complete and saved.
Render Method that writes out the html markup associated with the control.
Unload Do final cleanup, such as closing files or database connections
117

118
 Session State
 Cache

 e.g.,
 Presentation Layer
 Business Logic Layer
 Data Layer

120








 HttpApplicationState



121

 Application HttpApplicationState
 Page

 Page.Cache
 Session













State Management
ViewState Stored in _VIEWSTATE hidden field
ControlState Override if ViewState is turned off on the Page
HiddenField control
Cookies Add cookie data to the Cookies collection in the HttpResponse object
Query strings
Application State HttpApplicationState
Session State
Profiles SqlProfileProvider
122

 ViewState






123

125
 ViewState
 Passed to the client in _ViewState hidden field in the
Response message
 Returned to the Server on a postback Request
 Inspect using

 View html Source on the client
 3rd party tools like Fiddler
 Be careful of
 Data bound controls (GridView, etc.)
 Large TreeViews, DropDownLists, etc.

127
 Page.Application

Application.Lock();
Application["PageRequestCount"] =
((int)Application["PageRequestCount"])+1;
Application.UnLock();

128
ASP.NET Session state
 HttpContext.Session
 Data associated with a logical sequence of Requests
that need to be persisted across interactions
 Unique session IDs are passed along automatically with
each Request as either cookies or embedded in the
URL
 Session data is stored in a Dictionary collection,
allowing individual session state variables to be
accessed directly by Key name
 Three external storage options
 InProc
 StateServer
 SQL Server
 Timeout management

129
 HttpContext.Session
 InProc option provides fastest service, but does
not permit access to Session data from a
different process in a Web garden application
or a different Web server in a cluster
 Using alternatives to InProc session storage has
significant performance implications
 StateServer
 SQL Server
 Custom provider
 Measure impact using the IIS
RequestNotification events
 e.g., AcquireRequestState, PostAcquireRequestState

130
Session State options
 InProc
 StateServer



 SQLServer




132
Session State options
 InstallSqlState.sql InstallPersistSqlState.sql
 sessionState


 Frequency of re-use
 Cost to re-generate
 Size

Max(Cache Hit %)
Min(Space x Time)
133






diminishing returns
134
0 25 50 75 100
Cache
Size
Cache Hit %



 IIS kernel mode Cache
 ASP.NET Application cache
 ASP.NET Page cache






135






 effective
 Note: difficult complex

136

137

 HttpContext.Cache
 Page.Cache


 @ OutputCache
 VaryByParam


 <PartialCaching>

 Substitution.MethodName
 HttpResponse.WriteSubstitution


138



 frequency of use x object size




 Underused


140
 Cache Dictionary
 IEnumerable





141

 Add



 Insert
 Get
 Remove

142







 AggregateCacheDependency
 SqlCacheDependency

 http://msdn.microsoft.com/en-
us/library/system.web.caching.sqlcachedependency.a
spx

143
 effectiveness
 Cache API Entries
 Cache API Hits
 Cache API Misses
 Cache API Hit Ratio
 Cache API Turnover Rate

144

public void RemovedCallback
(String k, Object v,
CacheItemRemovedReason r){
}
 Examine the CacheItemRemovedReason







 InstrumentedCacheValue
// unfortunately, it is a sealed class; otherwise…
// private Int64 _refcount;
// private DateTime _lastreftimestamp;
145

146

 Count
 EffectivePrivateBytesLimit

 EffectivePercentagePhysicalMemoryLimit


disableMemoryCollection
disableExpiration
privateBytesLimit
percentagePhysicalMemoryUsedLimit privateBytesPollTime="HH:MM:SS"

147


 @ OutputCache
 Duration
 VaryByParam
 Location







149

 Response.Cache.SetCacheability
HttpCacheability







150
 effectiveness
 Cache Entries
 Cache Hits
 Cache Misses
 Cache Hit Ratio
 Cache Turnover Rate

151
 When page caching will not work due to
dynamic content


[PartialCaching(120)] public partial class
CachedControl :
System.Web.UI.UserControl
// Class Code

 Banner ads, etc.



 that did
not change
 AJAX
Javascript asynchronous HTTP
requests


152



 ICallbackHandler





<asp:ScriptManager EnablePartialRendering="True“ />




 153


 XMLHttpRequest
xmlhttp.open('GET', url, true);
xmlhttp.onreadystatechange = AsyncUpdateEvent;
xmlhttp.send(null);
154

 UpdatePanel





 Triggers
 ChildrenAsTriggers
 “ AJAX Application Architecture, Part 1
155


 Sys.Application.Init



 Sys.Application.Load

UpdatePanel

156

 PageRequestManager



 pageLoading
 pageLoaded

 endReques

157



 ICallbackEventHandler
 RaiseCallbackEvent
 GetCallbackResul





 Implementing Client Callbacks Programmatically Without Postbacks
in ASP.NET Web Pages 158

 REST
 AJAX library wraps asynchronous WCF service endpoints

 Try the new Asynchronous WCF HTTP Module/Handler
for long running Requests
 New in VS 2008 SP1
 Releases the IIS worker thread immediately after processing

 See Wenlong Dong’s blog for details

 http://msdn.microsoft.com/en-us/library/cc907912.aspx
159




 WebHttpBinding


160

 Binding
161
Binding Response Time (msecs) Throughput
wsHttpBinding 1300 1200
basicHttpBinding 1150 1800
netTcpBinding 400 5100
netNamedPipeBinding 280 7000
0
2000
4000
6000
8000
Throughput
Source: Resnik, Crane, & Bowen,
Essential Windows
Communication Foundation,
Addison Wesley, 2008. Chapter 4.
See also, “A Performance Comparison of
Windows Communication Foundation (WCF)
with Existing Distributed Communication
Technologies”
Caution: your
mileage will vary.




[ServiceContract(Session = true)]
interface IMyContract {...}
[ServiceBehavior(InstanceContextMode =
InstanceContextMode.PerSession)]
class MyService : IMyContract {...}
<netTcpBinding>
<binding name="TCPSession">
<reliableSession enabled="true"
inactivityTimeout="00:25:00"/>
</binding>
</netTcpBinding>




 in the order in which
they are received

conversation







[ServiceBehavior(
InstanceContextMode=InstanceContextMode.Single,
ConcurrencyMode = ConcurrencyMode.Multiple )]
class MySingleton : IMyContract, IDisposable


 ConcurrencyMode.Multiple
 requires locking and synchronization of shared state
data
 ConcurrencyMode.Reentrant

 serviceThrottling
<behavior name="Throttled">
<serviceThrottling
maxConcurrentCalls="100"
maxConcurrentSessions=“30"
maxConcurrentInstances=“20" />

performanceCounters
 Mark Friedman
Architect
Developer Division


<system.serviceModel>
<diagnostics performanceCounters="All"></diagnostics>
But
<diagnostics
performanceCounters=“ServiceOnly"></diagnostics>
is recommended (less overhead)
167

169

 “Using Service Trace Viewer for Viewing
Correlated Traces and Troubleshooting”
 Essential Windows
Communication Foundation


new "http://localhost:8000/HelloService"
string "http://localhost:8000/HelloService/MyService"
using new
new
"Press <enter> to terminate service"

172








 Measure:



 173







 TCP Window Size



without




175

Best Practices for Improving Page Load Time




 (serialized)









181











 184





 Get
Request


185

Perf by design

More Related Content

What's hot (17)

Similar to Perf by design (20)

More from Tess Ferrandez (12)

Recently uploaded (20)

Perf by design