Hi Oliver, I also ran into this problem with some debian distributions. I've filed it, personally, as a java issue with linux (since on my home PC where I have ubuntu and windows7, on windows one jmeter instance runs up to 6000 threads on a single process before I run out of CPU, while on linux the response times are much more dispersed even with only 300 threads - the jmeter script used wasn't dependent on external applications).
For actual load tests, on dual quad core machines with 12-24Gb of RAM, I ran tests with up to 8-10 JMeter instances. The criteria is not to exceed 200-400 threads per instance of JMeter. The problem is that the high throughput on each test client and each JMeter instance has to handle, so I had to tweak a lot about the OS user that runs the test, but the system itself is optimised to run actual web applications, so, my job was easy. I didn't need to scale it vertically more than this even if there was more room for it, since I have enough machines to distribute the test horizontally as well and until a few months ago I could still put down the system under test with only my desktop station. Although its nice to see how much you can take out of each single system, I would recommend to play it safe always and use more machines when this question arises. But pleae share the juicy details once you have them. Regards, Adrian S On Tue, Sep 13, 2011 at 1:56 AM, Oliver Lloyd <[email protected]>wrote: > Thanks guys, this is really useful stuff. I'm still being impressed by what > can be done with this tool. > > I'm play with this over the next few days and report back. > > ----- > http://www.http503.com/ > -- > View this message in context: > http://jmeter.512774.n5.nabble.com/Vertically-Scaling-JMeter-tp4792729p4796310.html > Sent from the JMeter - User mailing list archive at Nabble.com. > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > >
