Hi,
we're in a dispute with our SAN supplier and need to benchmark disk I/O over iSCSI and NFS (want to test NFS as I am quite sure the performance issues come from the iSCSI stack).
To do this I first tested the SAN with iozone (www.iozone.org) on the SAN itself. The SAN is running linux so it was fairly easy to format one of the raid sets and run iozone against it. Then exported the partition over iSCSI and formatted it with VMFS from ESX and created a linux VM on it.
Running iozone in the linux VM however has extremely surprising results. Test takes over 2 hours in the VM, yet it yield results that are stunning especially in the beginning with small files. It reports read/write rates of over 2GB/s. The NIC that serves as the iSCSI initiator however lists 6.6MB/s max and usually doesn't go over around 1.5MB/s. As the test gets to the larger files it gets down till around 8-10MB/s which still exceeds the graphs on NIC traffic by far.
Even in the section with small files, where iozone returns read/write's of 2GB/s and up this doesn't make sense. If I run the same iozone benchmark on my tmpfs (in memory filesystem) on my laptop I get up to about 1.5GB/s. What is surprising in this is that the entire test runs in less than a minute in tmpfs on my laptop (consumer grade hardware, servers will have much larger memory busses etc.) and goes to the next test really fast, whilst in vmware where results are often twice as high it will take much longer to go to the next test which doesn't make sense by looking at the results at all.
Basically I'm looking for a way to properly test iSCSI / disk I/O performance. Currently don't care about the rest. Any tips on using iozone would be welcome as well, these statistics don't make sense whatsoever. Looking at vmmark, but it seems much more app oriented than storage oriented.
Currently use a fairly simple iozone line: iozone -a -b /path/to/excel/file -f /path/to/tempfileforbench
In order to be able to compare results to bare SAN performance a tool that runs under linux would be prefered. Also tried running windows as iscsi initiator against the box, but it just gives headaches (as in test takes ages to complete averaging on 1.5-2MBps, I can send it over the internet way faster :smileygrin:).
TIA
