I have my 7110 serving a NFS file share for XenServer VMs and an iSCSI LUN to my backup exec server and I am noticing some poor performance on it. I recently migrated my Exchange 2003 box over to an Exchange 2007 VM and since then I have observed lag within all VM using the 7110 storage. I've got NIC0 set as my management NIC and NICs 1-3 aggregated into one link and plugged into an HP 2820 switch. The 3 gig ports on the HP as set as trunks. Prior to the Exchange migration I had been running a number of VMs on this setup and it had been running quite smoothly. I have a feeling I have the networking portion of this buggered somehow. Xenserver cannot use Jumbo frames so I can't try enabling it to see if it helps out.

Any ideas on how to improve performance?

 

 

Hi,

Exchange 200x drives significant small I/O and your probably seeing an IOPS bottleneck as a result. A mirrored configuration would be the best format for frequent small I/O database behaviors. As well if the datastore is located on the NFS share keep your database I/O size smaller e.g. 8K.

Turn off compression id you have it on.

You should add up the number of drives in your datastore volumes array and calculate the maximum IOPS that the volume can sustain e.g.

8 disk spindles * 125 IOPS per spindle = 1000 IOPS

Then use your analytics BUI to determine what the actual levels are. If they exceed or are close to the max then your options are very limited e.g cache mode.

Regards,

Mike

http://blog.laspina.ca/

 

 

Mike,

Thanks for the reply. I used the default project settings so the "db record size" is 128k. All my VMs are running off this single NFS share, is it ok to change this to 8k? If so, should I shut them all down while making the change?

I've got all 14 disks in a RAID 6 array. Am I able to remove disks from this array to make a single mirrored array? Would a RAID10 setup also be ok to use as I need larger space than my 300gb drives have?

I've had a poke around the analytics on the box and I think I have the found the form you are talking about. Is it "Disk: I/O operations per second broken down by disk"? If so, during regular hours (ie non-backup intervals) the box is well under 1000 OPS, usually in the 500-700 range. Although there are really short peaks to over 15,000+ OPS. During the backup of the exchange info store it sits at around 1500 OPS. This is done at midnight so should have little impact on users. I should mention that over the weekend the backup time period halved, from over 6 hours to 3 hours. The OPS during this time also dropped, from 3k-9k down to the 1500 I am seeing this week.

This sudden change just adds to my confusion.