It all boils down to what instance type you use. The cheaper ones aren't usually a good choice if you are testing concurrency and performance issues. If you are trying to establish peak performance numbers and you are doing distributed tests between hosts, m1.large is ok sometimes as long as the code isn't CPU intensive, but I usually go with c1.xlarge for the best mix of CPU and I/O. Twice as expensive, but it takes less time to reproduce obscure concurrency and network I/O issues. I have also had concurrency bugs that were difficult to reproduce on c1.xlarge that would show up within minutes on cc1.4xlarge (at $1.30/hr per machine, the sooner the better). Also, the higher end machines are less susceptible to slow down from other aws tenants.
So YMMV, but don't skimp on the machine type if you really want to push your code. HTH. Chris -- Chris Custine My Blog :: http://blog.organicelement.com On Fri, Dec 16, 2011 at 12:30, Alex Karasulu <[email protected]> wrote: > Hi guys, > > I was just thinking from past experience that we got some weird > results when using virtual machines to test MINA performance issues > dealing with concurrency. The presence of a hypervisor seemed to mess > things up. > > I don't know if using Amazon will be a good option for a perf testing > environment. > > Thoughts? > > -- > Best Regards, > -- Alex >
