So I’ve been playing around with OpenFiler in the dev environment I’ve cobbled together all made up of systems all nearing or already out of warranty support.
What I wanted to test was whether or not there was a difference between disk performance for an ESXi 5 VM with disk drives having a volume assigned to their own iSCSI target or could I map all the volumes to one iSCSI target and call it good.
In this situation, I have a Dell PowerEdge 2950 with OpenFiler installed along with 6 SATA drives in a hardware controlled RAID5. (PERC6i, I think)
Summary of the settings:
RAID 5 – 6 SATA 7.2k drives
2 iSCSI connections from Broadcom NetXtreme 5708 NIC
2 iSCSI connections from Intel 82576 NIC
Openfiler installed (latest distro)
Then I have a Dell PowerEdge 2950 with ESXi 5 installed.
Summary of the settings:
3 iSCSI connections from Intel 82576 NIC
ESXi 5 installed
So I added the Volume Group, added 2 separate volumes each worth 500GB. Then, after starting the iSCSI Target Service, I added a single iSCSI target and mapped both volumes.
Now I get into the ESXi host and add the IPs from the OpenFiler connection to the iSCSI software initiator and rescan the HBA. I configured all 3 iSCSI connections on the host in a Round Robin connection to the devices.
Part of the fun I found was that the switch I’ve obtained (an out of warranty 10/100/1000 24 port HP) does not happen to support jumbo frames. So all of this is ran with the MTU set at 1500.
So I take a VM already running on local storage of the ESXi host, add 1 2GB drive from each of the newly added volumes.
I opened up IOMeter, configured 2 workers (one for each volume) to run all the tests for 5 minutes and let them rip.
Surprisingly, I found that it performed at 1158.879 IOPS at 14.87 MBps total. Volume 1 was 580.365 IOPS and 7.437 MBps and Volume 2 was at 578.514 IOPS and 7.432 MBps.
Now I removed the drives from the VM, unmapped both volumes from OpenFiler and deleted the iSCSI Target. I created 2 new iSCSI targets and assigned one volume to each target and rescanned the HBAs on the ESXi host. Then I added 2GB drives (1 from each volume) to the VM again and ran the same test.
Another surprise, it was slower. The separate targets performed together for 1101.91 IOPS and 14.156 MBps. Volume 1 was 547.439 IOPS and 7.039 MBps and Volume 2 was at 554.471 IOPS and 7.118 MBps.
While the difference isn’t drastic, it’s enough to not bother configuring separate iSCSI targets when dealing with OpenFiler environments.