Re: [OpenSTA-users] Performance testing with Open STA
Brought to you by:
dansut
|
From: Bernie V. <Ber...@iP...> - 2007-03-09 22:13:11
|
<Chris wrote> > You are right there appears to be a bottleneck somewhere as the system Monitoring CPU utilization on the servers under test (SUTs) would be a good start. If you have Hyperthreading capable CPUs on the SUT(s), turn HT off (it distorts CPU accounting and can cause the OS to under-report CPU utilization by 40-50% under most workloads. This can only be done at the console at boot time on the servers I am familiar with). > cannot support 5 or 10 users using the system simultaneously. The > developers informed me that this is not a ralistic situation as there is > no way 5 or 10 users will hit the same buttons and perform the same > function at the same time. Thats a prettylow number based on what you say in the following paragraph. But I have no idea of the system configuration, how demanding the transactions are, what if anything is happening on the systems you are testing, nor the network bandwidth between the OpenSTA load generator and the SUT(s). All these are worth considering and understanding to put the "5-10" user limit into context. > I am currently running a test for 5 virtual users, and ramping up the > users such that 1 user is added every 30 seconds with a 10 seconds delay > between batches and the test is run for an hour. The recorded test takes > about 5 minutes to run (Login, Create order Logout)for a virtual user > and this emulates the real life situation. The problem im having now is, > although there are no error in the application's logs, there are time > out errors in the error log of OpenSTA, the orders appears to be > correctly created in the database. Could this be OpenSTA timing out, I > understand there is a timeout of 1 minute used by OpenSTA, will there be > any effect on my application if this timeout is increased? As a rule of thumb, I would set my ramp up period to be no less then the longest running script, in your case, 5 minutes. I would ignore response times during the ramp up and instead watch what happens to response times and server resource utilization (disks, network, CPU, memory,...) in the 10 or 15 minutes following the end of the rampup period. Also, increase the OpenSTA timeout to something large, say 300000 (unites are milliseconds by default, so this will give you up 5 minute timeout ) so you can see the true response times. Someone earlier on mentioned how to do that. Its a setting in an .ini file. Just Google OpenSTA timeout if you can't find it. -Bernie (www.iPerformax.com) |