On Sun, April 21, 2013 17:09, Tanstaafl wrote: > On 2013-04-21 5:47 AM, J. Roeleveld <[email protected]> wrote: >> On Sat, April 20, 2013 17:38, Jarry wrote: >>> Problem of virtualized filesystem is not that it is virtualized, >>> but that it is located on datastore with more virtual systems, >>> all of them competing for the same i/o. *That* is the bottleneck. >>> If you switch reiser for xfs or btrfs, you might win (or loose) >>> a few %. If you optimize your esxi-datastore design, you might >>> win much more than what you have ever dreamed of. >> >> If the underlying I/O is fast enough with low seek-times and high >> throughput, that handling multiple VMs using a lot of disk I/O >> simultaneously isn't a problem. Provided the Host has sufficient >> resources >> (think memory and dedicated CPU) to handle it. > > My host specs: > > Dual AMD Opteron 4180 (6-core, 2.6Ghz) > 128GB RAM > 2x internal SSDs in RAID1 for Host OS > 6x 300G SAS 6Gb 15k hard drives in RAID10 for Guest OSs
Sounds like a nice machine for testing :) > I allocate each Guest 1 virtual CPU with 2 cores Do you limit the Guest to use any of 2 specific cores? Or are you giving 2 vCPUs to each Guest? >> A decent hardware raid-controller with multiple disks running in a >> higher >> raid version is cheaper then the same storage capacity in SSDs. > > Yep... I toyed with the idea of SSDs, but the cost was considerably more > as compared to even these SAS drives... I am planning on using SSDs when getting new desktops, but for servers I prefer spinning disks. They're higher capacity and cheaper. For speed, I just put a bunch of them together with hardware raid. -- Joost
