CODEWITHSUNDEEP
sharepointvisual-studio-2008load-testingwebtest
ravikanth
ravikanth

Reputation: 25820

Setting up a basic load test

I have a very basic load test question.

I am running a load test using VSTS 2008 and I have test rig with controller + 10 agents. This load test is against a SharePoint farm I have. My goal of the load test is to find out the resource utilization on web+app+db tiers of my farm for any given load scenario. An example of a load scenario is

Usage profile: Average collaboration (as defined by SCCP)

User Load: 500 (using step load pattern=a step of 50 every 2 mins and a warm up time of 2mins for every step)

Think time: 0

Load duration: 8hrs

Now, the question is: Is it fair to expect that metrics like Requests/sec, %processor time on web front end / App / DB, Test/sec, and etc become flat or enter a steady state at one point in time during the load test. Like I said, the goal is not to create a bottleneck but to only measure the utilization of resources by the above load profile.

I am asking this question because I see something different. At one point in the load test, requests/sec becomes more or less flat. But processor utilization on the web/DB servers keeps increasing. After digging through the data a bit, I see that "tests running" counter also steadily increased over time. So, if I run the load test for more than 8hrs, %processor may go up further. This way, I don't know what to consider as the load excreted by the load profile.

What does this "tests running" counter really signify? How is this different from tests/sec?

Another question is: how can I find out why "tests running" counter shows an increase overtime?

Thanks for your time

Upvotes: 3

Views: 533

Answers (1)

Ariel
Ariel

Reputation: 5830

With no more info than provided, I guess it may have something to do with requests queueing and worker thread's pool.

Worker pool has a limited number of threads to use to serve requests. Once that's reached (which can after a while, because of front end performing DB requests and waiting for them, very likely not asynchronously) then requests get queued (I think that's your "tests running" counter increase).

Why more CPU gets used? Again, you need to dig into scenario's data, but one guess is that context switching and memory swaps and allocations may have something to do with that.

Take this lightly, as without more data, nothing can be said positively.

Upvotes: 2

Related Questions