We have a cluster of four ESX 3.0.2 servers connected to an HP MSA1000 SAN. Running a Windows Server 2003 virtual machine with SQL Server 2005, we are noticing slow performance compared to the physical machine it was converted from running large SQL queries. A query that takes 3-4 minutes on the physical machine takes 10-12 minutes on virtual. The performance issue appears to be disk related, as CPU, RAM and network performance counters show no problems. I have tried having the database files stored on a virtual disk on VMFS on the same LUN as the system disk, a different LUN to the system disk, and on an RDM. No noticeable difference in performance between those (RDM is maybe marginally faster, as you'd expect). I created a datastore on the local scsi disk on the ESX server (HP Proliant DL380, SmartRaid with two 137GB disks in a mirror) - and performance is much better, only marginally slower than the physical server (also an identical Proliant DL380).
So the issue appears to be with the shared storage. I have checked the firmware and driver versions and they seem current. However I have noticed something which I wonder if it's significant. The HBAs in the ESX servers are HP FC2243 (Emulex LP11000) 4Gbps fibre channel cards. We have two per server, each connected to a separate HP 4/10 FC switch, which connect to two MSA1000 controllers. The HBAs are 4Gbps, the switches are 4Gbps, but the MSA1000 controllers are 2Gbps. According to cat /proc/scsi/lpfc/0 the HBAs are connected at 4Gbps to the switch and are running firmware 2.10A7. The connection from the switch to the MSA1000 must be 2Gbps.
Could this be a factor in the slower performance? Is it possible to set the cards to 2Gbps perhaps to rule this out? Or does anyone have any other ideas...?
Thanks.
Rich
