Cuyahoga performance

In the past different people have asked information about how many users Cuyahoga can handle. Finally we have some information about this subject that might be useful.

For the 1.5.0 release we have done some performance tests to verify that the new architecture with Castle performs just as good as the 1.0.x release and to make sure there are no memory leaks.

All tests were performed with Microsoft Application Center Test and the server was a dual core Pentium D 2.8 Ghz with 2 GB RAM with Windows 2003 Server, IIS 6 and SQL Server Express 2005. The test site was a copy of this very site, (25 pages (forum excluded) and 800 registered users). For most sections, output cache is set to 60 seconds and the second level cache is set to 30 seconds.

So here are the results:

Cuyahoga 1.0.1
About 230-260 requests per second with 95% CPU load. This includes static content, but most requests were aspx pages.
Memory usage for the w3wp.exe process was stable at 50MB.

Cuyahoga 1.5.0
About 280-400 requests per second with 90% CPU load.
Memory usage for the w3wp.exe process was stable 61 MB.

Remarks:
- During the tests, there was almost no load on the database.
- The differences in requests per second for 1.5.0 were caused by modules that leverage Castle's automatic transaction management (Articles and Downloads). Creating one of these modules almost doubled the page execution time.
- We did have a memory leak in the 1.5.0 version. This was because we registered all aspx pages in the Castle container and it seemed that those pages were not released properly. After removing registering the aspx pages in the container, memory usage was stable.

In conclusion we can say that Cuyahoga 1.5.0 now performs at least as good as Cuyahoga 1.0.x. Implementing the Castle container in a sensible manner didn't hurt the performance. All credits to the Castle guys for building these brilliant tools!