On Sunday, 20 March 2011, Kevin Chadwick <ma1l1i...@yahoo.co.uk> wrote: > On Sat, 19 Mar 2011 08:29:13 -0700 > Ben Calvert wrote: > >> On Mar 19, 2011, at 7:49 AM, Kevin Chadwick wrote: >> >> > On Fri, 18 Mar 2011 16:58:59 +0000 >> > Kevin Chadwick wrote: >> > >> >> I do get a fair increase in cpu usage for a disk at full speed disk with >> >> vnd but it's acceptable. Have people already done cpu usage and >> >> transfer speed comparisons to save me further tests. >> > >> > Well I was about to run a comparison test on vmware and I'm well >> > confused unless it's a strange vmware bug or maybe the dynamic size disk >> > mechanism. I might have to pull out a box. >> >> Why do people do this? >> >> when you're running more than one OS at a time, there's no way to control >> what's running on the other system(s) and interfering with the process you're >> testing. or what vmware subsystem is thrashing around and creating overhead. >> > > Surely that would affect both partitons inconsistently. Of course I see > the point but if you get a good idea of it quickly especially if there's > a big difference and have lots to do and bear it in mind and can control > your host (ignoring vmware) then it should be fine. I'm still confused > why it isn't but I could probably just use one partition to reinforce > the results at marcos link. I am definately far more worried about > vmware bugs than my host systems doings when testing, though it's > usually just configs and so doesn't matter unless something crashes. > One of the relayd socket engines started crashing recently but it's > stopped now and obviously I'd test that on a real system if it reoccurs. > > p.s. Thanks marco
Bear in mind that as great as the benefits to virtualization can be, or appear to be, it introduces some quirky edge cases that make it especially bad for benchmarking. I've seen benchmarks performed in VMware which return results that are far far faster than when the benchmark/app/OS is running on the bare metal. And others where the realworld performance is bad, but the benchmark numbers are good, due to timing lies. If something interesting came out of a benchmark under VMware, I'd be wondering if it was significant or just a virtualization quirk. But worse still, performance issues can also be masked and thus missed. Unknowns that invalidate benchmarking period for me. Shane
