Using Process Monitor to Troubleshoot Internet Explorer 7 Performance Issues

[Added 2007-01-18: Fix Available for Performance Problems with Internet Explorer 7's Phishing Filter ...]

Previously, I wrote about sluggish behavior with Internet Explorer 7. I had used Process Explorer to help pinpoint the cause of the sluggishness - in this case, it was Internet Explorer 7's (anti)phishing filter. I could also have used a relatively new tool from Microsoft's Windows Sysinternals - Process Monitor.

In Process Monitor, it is easy to get inundated with all of the data that the tool collects. Filters are very critical to enabling one to find the desired information, and the implementation of filters in Process Monitor is top notch. For this exercise, after starting Process Monitor and checking "Generate Profiling Events" on the "Options" menu, I captured events while exercising the web application with the Phishing Filter set to "Turn off automatic website checking" and "Enabled" for the Security zone the web application was in. After capturing events for a minute or so, I set the following filters:
-Process Name is iexplore.exe then Include
-Event Class is Registry then Exclude
-Event Class is File System then Exclude
-Event Class is Process then Exclude

This displayed all "profiling" events for the iexplore.exe process. I double-clicked an event to bring up the "Event Properties" dialog and clicked on the "Stack" tab, which had the following information:

kernel32.dll!WaitForSingleObject + 0x12
ole32.dll!GetToSTA + 0x6f
ole32.dll!CRpcChannelBuffer::SwitchAptAndDispatchCall + 0xf6
ole32.dll!CRpcChannelBuffer::SendReceive2 + 0xb9
ole32.dll!CAptRpcChnl::SendReceive + 0xab
ole32.dll!CCtxComChnl::SendReceive + 0x113
RPCRT4.dll!NdrProxySendReceive + 0x43
RPCRT4.dll!NdrClientCall2 + 0x1fa
OLEAUT32.dll!IDispatch_RemoteInvoke_Proxy + 0x1b
OLEAUT32.dll!IDispatch_Invoke_Proxy + 0xb6
ieapfltr.dll!ATL::CComPtr::GetProperty + 0x56
ieapfltr.dll!FieldContainer::VisitInput + 0x1b9
ieapfltr.dll!FieldContainer::VisitAllElements + 0x21d
ieapfltr.dll!FieldContainer::ExtractFieldCount + 0x10e
ieapfltr.dll!FieldContainer::InitFieldCount + 0x9
ieapfltr.dll!PageDetails::Init + 0x315
ieapfltr.dll!PageDetails::Factory + 0x59
ieapfltr.dll!HeuristicsFeatures::InnerExecute + 0x15b
ieapfltr.dll!HeuristicsFeatures::Execute + 0x55
ieapfltr.dll!ProcessingThread::RunPageAnalysis + 0x1b4
ieapfltr.dll!ProcessingThread::RunUrlAndPageAnalysis + 0xdb
ieapfltr.dll!ProcessingThread::Analyze + 0xd3
ieapfltr.dll!ProcessingThread::AnalyzeFrame + 0x249
ieapfltr.dll!ProcessingThread::EnumerateFrames + 0x2e5
ieapfltr.dll!ProcessingThread::EnumerateFrames + 0x249
ieapfltr.dll!ProcessingThread::Evaluate + 0x1ec
ieapfltr.dll!ProcessingThread::Execute + 0x78
ieapfltr.dll!ProcessingThread::Process + 0x24e
ieapfltr.dll!ProcessingThread::Start + 0x72
ieapfltr.dll!Evaluator::ContinueProcessing + 0x21f
ieapfltr.dll!Evaluator::ContinueProcessingWrapper + 0x21
ntdll.dll!RtlpWorkerCallout + 0x70
ntdll.dll!RtlpExecuteWorkerRequest + 0x1a
ntdll.dll!RtlpApcCallout + 0x11
ntdll.dll!RtlpWorkerThread + 0x87
kernel32.dll!BaseThreadStart + 0x37

With this information from Process Monitor, one could come to a similar conclusion - the Phishing Filter in Internet Explorer 7 seems to cause Internet Explorer 7's performance to degrade in certain environments.



The Case of the Sluggish Internet Explorer 7

[Added 2007-01-18: Fix Available for Performance Problems with Internet Explorer 7's Phishing Filter ...]

I like Internet Explorer 7. I've installed it many, many times and on
many systems. I've never had a problem with it. That is, until I had to spend some time working with a web-based application on an Intranet. I had to go through several iterations of repetetive steps in this application. It was the kind of work where it would have been more fun to write a program to achieve the end result, but it would probably have taken more time to write the program than it would to go through the tedious process. That, and the fact that I didn't immediately have access to some of the information that would be required to write the app, prevented me from taking the fun route. So I was stuck copying, clicking, pasting, and... WAITING.

The problem wasn't the responsiveness of the web server - the problem was localized to my system. As I was working on a laptop, I could hear the fan kick into high gear as CPU utilization hit 100%... and stayed there. When I paused for a bit, the CPU usage went back down. So, interacting with the web application was causing the behavior. The task took about 90 minutes to complete, and when I was done, Process Explorer showed iexplore.exe as having used nearly 90 minutes of CPU time. I didn't have this problem with Internet Explorer 6! In communicating with the vendor of the web application, they indicated that they hadn't had problems of this sort in any of their experiences with IE7 or with other customers.

I surmised that there was likely a setting in Internet Explorer 7 that was affecting the performance. But IE7 has no shortage of settings, and to try each one was not a task that I wanted to undertake. I fired up IE7, loaded the web-based app, and started working. Then, while the CPU was taxed, I went to Process Explorer, hit the properties of the iexplore.exe process, and checked the "Threads" tab. There were 89 threads, most having a start address of "ndtll.dll!RtlpWorkerThread", and all vying for CPU time. It appeared that a new thread was created for each request that was made, which seems rather "wasteful". At any rate, RtlpWorkerThread is a private "run-time library" worker thread threadproc function, presumably the threadproc used when one uses the Thread Pool API. The stack of one of these threads at the point that I captured it (obtained with Process Explorer and properly configured debugging symbols) is rather deep:


As you can see, after the thread pool plumbing is out of the way, the first function in the stack is ieapfltr.dll!Evaluator::ContinueProcessingWrapper - a function in ieapfltr.dll. ieapfltr.dll describes itself as "Microsoft Phishing Filter", though I suspect IEAP is an acronym for Internet Explorer Anti-Phishing. At any rate, it certainly seemed that the settings surrounding the (anti)phishing filter would be a good place to start.

I went into "Internet Options" in the Control Panel (Start Run inetcpl.cpl) and hit the "Advanced" tab. Toward the bottom was a setting for the Phishing Filter where there were three options:
-Disable Phishing Filter
-Turn off automatic website checking
-Turn on automatic website checking

Mine was set to "Turn off automatic website checking". I changed it to "Disable Phishing Filter" and hit OK, and re-tried the web application. The application was responsive as ever, and the "excess" threads that were previously being created were nowhere to be seen. Changing the setting back to "Turn off automatic website checking", and hitting the web application again caused iexplore.exe to consume as much of the CPU as it could. It appeared that I found the cause of the sluggish performance.

It is worth noting that in the "Internet Options", on the Security tab, each Internet Explorer Zone has its own setting for the Phishing Filter. If one selects a Zone, and clicks the "Custom level..." button, about 75% of the way down is a setting for "Use Phishing Filter" that has 2 options - Disable or Enable. The setting on the Advanced tab overrides this - if the Security zone setting is set to "Enable" and the Advanced setting is set to "Disable Phishing Filter", the Phishing Filter is disabled for all zones. However, if the Advanced setting is set to "Turn off automatic website checking" or "Turn on automatic website checking", one can exercise more granular control over each Security zone by choosing to enable / disable the filter where it makes sense.



Driver Framework Resources

Finally got through watching some good overview videos on MSDN's Channel 9.

Doron Holan talks about the Kernel Mode Driver Framework (KMDF) in this Channel 9 Video Segment. Check out the state machine diagrams!

Peter Wieland goes over the User Mode Driver Framework (UMDF) in this video segment. There's even some discussion about writing a driver with managed code (C# / VB.NET).



How DOES this warrant a KB article?

In How does this warrant a KB article?, I pondered the existence of a Microsoft KnowledgeBase article that discussed how to set the SmtpMail.SmtpServer property of the System.Web.SmtpMail class.

In my travels, I encountered a blog posting in the "The CDOs and CDONTS of Messaging Development" blog - MYTH: SmtpMail.SmtpServer.Insert(0,"") Actually Does Something. This posting is interesting because it's possible that it prompted the aforementioned KB article (922777: You receive an error message when you try to send an e-mail message by using the System.Web.Mail namespace in the .NET Framework 1.0). It's also interesting because it talks about the source of the infamous code - a Code Project article.



Brief Frustration With Global.asax

It was maddening, I tell you. I like to write C# code in a C# source file, not inline in a script tag. So I went about modifying Global.asax to allow me to do so:

<%@ Application language="C#" CodeBehind="Global.asax.cs" Inherits="Global" %>

I then went about defining a class called "Global" in Global.asax.cs. I added the code I needed and proceeded to build the ASP.NET app in Visual Studio 2005. But I got an error:

Global.asax(1): Build (web): Could not load type 'Global'.

Of course, I was perplexed. Why not? Why couldn't the type be loaded?
I thought for a bit, but then went about doing some more coding. Eventually, I had to address the problem, though. How? On a whim I placed Global.asax.cs in the App_Code ASP.NET folder. Once I did that, I was able to build the app.