ASP.NET Memory: If your application is in production… then why is debug=true

Statement

“Ensure that the debug="false" on the <compilation> element in the web.config file of each and every ASP.NET application on the server. The default during development is "true" and it is a common mistake to allow this development time setting to find its way onto production servers during deployment. You don't need it set to true in production and it often leads to memory overhead and inefficiencies.”

What problems does leaving debug=true cause?

There are three main differences between debug=true and debug=false:

  1. ASP.NET Timeouts
  2. Batch compilation
  3. Code optimization

ASP.NET Timeouts

When debug is set to true, asp.net requests will not time out. This is to allow you to debug with visual studio at your own pace without having to worry about the requests suddenly disappearing. Of course in a production environment timeouts are crucial to avoid for requests to be stuck indefinitely, so this is reason #1 to make sure debug is set to false when the application is deployed into production.

Batch compilation

In short, when debug=true, we don’t batch compile, when debug=false we do…

What does this mean?

When an aspx, asax, or ascx page is first requested it gets compiled into an assembly. This assembly has a name like 3ks0rnwz.dll or similar (8 characters) and stores the class for the actual ascx, asax, or aspx page (not the code behind). The assembly goes into a folder in the C:WINDOWSMicrosoft.NETFrameworkv1.1.4322Temporary ASP.NET Files with the same name as the application.

The code behind class gets compiled into the main dll for the assembly, and it along with all the other dlls in the applications bin directory get shadow copied to the Temporary ASP.NET files.

Back to the 3ks0rnwz.dll… If we have debug=true, we create one dll per aspx, asax, or ascx page and this dll is compiled in debug mode, so if you have 100 web pages, you will have 100 assemblies, and they are generated as the pages are requested.

If we instead have debug=false, we batch compile, which means that the first time you request any page in your application, we compile the whole batch into one big assembly. This is a truth with some modification. The user controls (ascx pages) are compiled into a separate assembly from the aspx pages and the aspx pages are compiled in groups based on what other files (read usercontrols) they include. The global.asax is also compiled separately. And batch compilation occurs on a directory bases, meaning that if your application has subdirectories, the subdirectories are compiled separately to avoid for example name clashes, as it is valid to have two aspx pages with the same name in different directories. But all in all, instead of 100 dlls, you might end up with 3 or 4.

Ok, big deal? It’s the same code so the size of the combined assemblies shouldn’t much differ from the size of the individual assemblies right? Truth is, there probably isn’t an enormous difference. But… and this is a big but… there is overhead for each dll, and if the dll is compiled in debug mode there is overhead for items needed for debugging, and … last but not least (in fact probably most important), the assemblies won’t be laid exactly side by side, so with a large number of assemblies you start fragmenting the virtual address space making it harder and harder to find large enough spaces to store the managed heaps, potentially causing out of memory exceptions.

One caution even if you have debug=false, is that if you go in and change something in one of your aspx pages, this page will have to be recompiled, but this doesn’t cause an appdomain reload so the whole application is not batch compiled again. This has the effect that the page will now get recompiled separately and get its own dll, so don’t change your aspx pages on a live server too often.

There is a setting in machine.config determining how many recompiles are allowed before the app domain restarts, by default it is set to 15, so after 15 recompilations the app domain will restart, just as it would if you touched the web.config or touched the bin directory.

Code optimization

In order to be able to step through code line by line the JITter can’t really optimize the code which means that your debug dlls will be less performant than if they were compiled in release mode.

So as you can probably figure, there is a large benefit to having debug=false in production…

How can you identify it in a memory dump?

To find out if any of the applications on your server run with debug=true you can run a nifty command in sos.dll called !finddebugtrue which will list out all applications where debug=true in the web.config, now how easy is thatJ

0:016> !finddebugtrue

Debug set to true for Runtime: 61b48dc, AppDomain: /MyDebugApplication

Debug set to true for Runtime: 1f50e6d8, AppDomain: /MemoryIssues

Total 16 HttpRuntime objects

And to find out if you forgot to compile some of your assemblies in release mode run !finddebugmodules

0:016> !finddebugmodules

Loading all modules.

Searching for modules built in debug mode...

MyDebugApplication.dll built debug

MemoryIssues.dll built debug

fl4sq-9i.dll built debug

wepr1t3k.dll built debug

r9ocxl4m.dll built debug

zmaozzfb.dll built debug

Done Seaching

The dlls above with weird 8 character names are the dlls generate when JITing the aspx pages, so they will go away when debug=false.

Oh, before I forget… when you change from debug=true to debug=false it is a good idea to clean out your Temporary ASP.NET files for this application so you don’t have some old junk in there causing it to still not batch compile.

In ASP.NET 2.0 there is a switch that can be turned on in machine.config that turns off all debug=true, so in 2.0 applications you can do this directly without worrying about finding out which applications do and don’t have it.

<configuration>

    <system.web>

          <deployment retail=”true”/>

    </system.web>

</configuration>

If you want some more goodies about debug=true, read ScottGu’s blog post about it https://weblogs.asp.net/scottgu/archive/2006/04/11/442448.aspx