01-24-2013 07:21 AM
I know this is a basic level question - still would like to post it since I see some strange behavior in my test result.
Does running a part of load on a localhost LG (both Controller and LG) have any impact on the application transaction response time observed during the test?
Theoritically I feel it should not be the case, but still want to confirm.
Env: LoadRunner 11
Total load: 50 Vusers
Load on local host: 10 Vusers
01-24-2013 11:08 PM
Depends on the RAM, protocol, as each vuser consumes memory and kind of application...if it's rich internet based...you might want to distribute the load on different hosts.
Running 50-VU on same host (LG+Controller) should impact on transaction response time.
04-03-2013 06:59 AM
Most definitely. How many users, the protocol used, even how long your tests run can cause an LG to influence the results of a test. That is why you should be monitoring your LG's to make sure they are not causing a problem.
For me I like to ensure that:
- CPU doesn't ever exceed 75% even spikes above that would be concerning
- Memory doesn't exceed 80% utilized.
When you run vusers on localhost you have to carve out a large chunk of RAM and CPU for use by the Controller, however this is no different than if you run any other software on an LG. The more programs you run on the system means less total virtual users.
When you are testing out 1, 2, 10 and depending on the protocol even 50 vusers to ensure that they can run with multiple users and multiple iterations then using LocalHost is not a problem, mostly because you are not really looking at the timings. However, when you need to bump up the number of vusers and the results are important, will be analyzed and decisions will be made based on the findings, I would strongly suggest using dedicated load generators and make sure that they are not interfering with the results of the test by being overloaded.
HP Premier Support - Technical Account Manager - ALM Products