Performance Testing

I have run some load tests against my 4.1SP3 website. It currently has about 300 pages and0 300 documents. I used a load testing servers to mimic up to 500 users being on the site at once.  It works by creating various scenarios of people going through the home page and clicking on random links.  I have a scenario for 1, 2, 3 and 4 clicks through the site.  Once a scenario is done by a user a new user is generated to repeat the random actions. It is possible to have 500+ virtual users going to the site every minute.

I have made some observations below and would like to know if anyone else has done the same thing or had the same experiences. This is running on a single quad-core Xeon 2.00Mhz, 8GB ram.  Our VMWare reports show that the peak CPU utilization was 50%, with a 25% average. This is highly unscientific. Your mileage may vary.

My average response times for most pages are 2-3 seconds with 350 user profiles, and 3-4 with 500, however pages that have image galleries sometime time out.

The TElerik.Web.UI.WebResources.axd files have the longest max response times, some being over 20 seconds.   Their average is in the top 15% of max load times, even for pages that have only been opened once in the test and are being built from cache as well as 3mb PDF files.

My site seemed to handle more requests per second and more total pages served over a 45 minutes span when the user limit is 300 than it does with 500.  400-500 users seem to slow the response times enough to make it less efficient.  Attached are two test summaries with tests done with a gradual increase from 250-500 users as well as one with an increase from 350-500 users.

I saw a statement about SF 3.7 that Telerik tested with two P4 processors and was able to serve 150,000 pages per hour without cache and 1,000,000 with cache. What was the response times? Has any testing of that nature been done on the 4.x versions?

Does 4.2 speed this up?  Our normal daylight hourly traffic is about 17,000 pageviews, with peaks of up to 26,000 in emergency situations.  Right now our SF install seems to operate in that range, but I would LOVE for it to run faster in those peak conditions.

Anthony


Anthony

Hello Anthony,

We have some official results that are currently featuring the following hardware and software configuration.

Hardware:
-    Intel Core 2 Duo E7300 @ 2.66GHz
-    4Gb RAM
-    MS Windows Server 2008 R2 Standard
-    IIS 7 and SQL Server 2008 R2

Site tested:
-    Sitefinity 4.x “Solar Flare Demo Site” (Solar Flare Free Template with Lorem Ipsum contents - www.sitefinity.com/.../solarflare.aspx)
-    Disabled cache
-    Single instance

Test:
-    100 concurrent users with no delay

Results:
-    Front-end output: 26 pages per second
-    Average page/time: 3.78 s
-    Application memory pool average for the front-end: 213Mb
-    Application memory pool average for the back-end: 500Mb
-    CPU usage: 100% (due to disabled cache)

We are constantly working on improving our performance until we achieve some outstanding performance. Performance improvement will be treated as our number one priority in the upcoming 6 to 8 months.

I cannot find the attachments you mention but you can directly send them on my corporate e-mail at hristo dot borisov at telerik dot com. Thank you for contacting us.

Regards,
Hristo Borisov
the Telerik team

Can you tell me your network configuration for this test?  Was SQL running on the same server as the website?  Were these tests done from inside the same network or from outside?

Hello Anthony,

The results are obtained with SQL Server and IIS being hosted on the same server and users hitting the websites within the inside network. We will make a new performance tests for the upcoming SP1 which will feature a dedicated SQL Server and a web server hosted most probably on Rackspace with tests performed from different machines outside the network.

Greetings,
Hristo Borisov
the Telerik team