We're considering moving from physical dedicated servers to leased virtual machines. I understand that a common problem with virtual machines is the performance of the SAN they are running from. As a quick test, I wanted to benchmark the performance of a few different virtual machines. I will test my own VM, as a control (no contention) and trial accounts with a couple of major providers like MS Azure.
I'm currently using HD Tune free, but I think I'm having a problem with caching. The performance seems low most of the time (slower than an old standard SATA drive) but then suddenly the test will spike to a really high level and messes up the results (as I'm only interested in raw performance - not cached performance). I also see this effect on one of the SCSI based machines (not a VM) I've tested on, so I can only assume that the SCSI controller and SAN controllers are using a large cache.
Is there a better way to test disk performance than using HD Tune, which is perhaps not subject to the problem if disk caching?
I have used the following to great effect in the past
Both specifically avoid caches. I would have a look at Iozone first and see if that fits your needs.
There's also IOMeter which is very widely used.