hi,
We've recently have some catastrophic failure on our exchange DAG node, and having long painful discussion with our storage vendor.
SCSI reset was issued most likely due to very high latency to disk, and had our log file corrupt.
Here's our current setup.
3-node Exchange 2010 DAG with 3 MDBs, active/passive/passive on each node.
Each node has 3 disks attached to them - OS, EDB and logs - all on single LSI SAS controller.
OS drive is thin provisioned
EDB drive is also thin provisioned - we had mailboxes migrated from old server into this first (~850Gb in total), We then archived a lot of emails out of these, there is 30%+ white space in the mailbox database. Hence, even though new emails arrive to the mailbox server, this should not be writing to new block, there is no need to zero out blocks hence no performance issue is expected - our storage vendor is insisting that thin provisioned disk will always run significantly slower than thick provisioned disk, even it's writing to existing blocks / reading from blocks. Any thought on this would be appreciated.
Log drive is thick provisioned. we had issue with backup at one stage which caused log drive to be filled out, so everything here should have been zeroed out already.
We have about 1200 mailboxes spread across 3 mdbs, and each mailbox send/receives less than 100 mails a day on average. Expected IOPs on this is less than 200 IOPs - http://technet.microsoft.com/en-us/library/ee832791(v=exchg.141).aspx
Our mailbox LUN is designed to guarantee minimum of 750 IOPs.
Our storage vendor engineer also insist we separate OS drive and data/log drives on different LSI SAS controller on VM.
I highly doubt this is necessary, as a VMware LSI SAS controller should not be a bottleneck for at least a few thousand IOPs. Anyone had LSI SAS controller being a bottleneck for the disk performance?
Any help on this would be very appreciated...
Thanks,
Mike