I would like to know what is the best iSCSI performance that users see in a virtualized environment. I have a setup with three AMD 2352 blade servers with Qlogic iSCSI HBAs and HP MSA2000i and Dell MD3000i SANs. I have tried SATA and SAS disks in both RAID10 and RAID5 configurations and the best throughput I can seem to get (according to CrystalDiskMark on Window 2003R2 and Window XP VMs) is about 75MB/s for sequential reads.
This is only about half of the performance I got from W2k3 on a PowerEdge 2550 with a PERC5/i RAID5 array. I was even using 15k SAS disks on my MD3000i while the PowerEdge just had 10k U320 disks.
Utilization on the GigE links to the MD3000i seems to top out at around 300Mbps each (I am using vSphere's "round-robin" multipathing).
While I realize that VM disk performance will probably never match that of a native OS, I would expect it to be a little closer. With the Qlogic iSCSI HBAs, processor utilization shouldn't be an issue. My GigE links are nowhere near max'ing out. I would think that if it were a physical disk limitation that I would not be seeing that kind of performance with the PowerEdge server. Is this an iSCSI limitation? I would think not since people seem to think a lot of iSCSI on 10GE links. Is it virtualization causing the bottleneck? Any thoughts would be appreciated!
I believe there is some configuration issue somewhere. Dont have too much time right now but i will come back later and respond again. However I am not sure how round-robin multipathing works.
Well was trying to get educated on round robin iscsi (keep in mind I am fairly new iscsi, i am in the middle of implementation of infrastructure, however I am sure we can figure out what is happening) So I did a quick search on round robin and vmware and came up to this page http://communities.vmware.com/thread/131295 now although it isnt the exact same setup the guy did some extensive tests and was able to get 180mb/s with multiple vm's running.
I personally didnt want to find numbers of some marketing study because I feel like they control their environments too much and their published results arent what we get in the real world. I will have to do some more research and maybe try to create a quick diagram of what the setup looks like.
Is there anyone else out there using vSphere? Cause im not that versed in Vmware's products. Any way I hope im helping rather than hindering. I am going to download CrystalDiskMark and see what I am getting. I will let you know once I do.
Team_DANS
5 Posts
390
0
Posted February 25th, 2010 10:00