Greetings,
Been using ESXi w/ Openfiler for a lab environment for over a yr and have recently set out to improve disk performance since adding a 4th WD RE4 to my Openfiler Adaptec 5805 and generally not being happy with performance on a per VM basis. Overall things run ok running upward of 20 lower demanding VMs at a time, however I'm wanting to know what ESXi is doing to "throttle" VMs disk performance, or so it seems, as my RAID5 and NICs got plenty of room past what a Windows VM is allowing.
Brief storage and test overview
Openfiler, 3 nics, seperate subnets, jumbo, write back, read cache, blockio, 4x 2TB SATA RAID5 256K stripe
ESXi, 3 nics + vmks, 3 paths active per lun RR, jumbo
Datastore in use for test 2TB 8MB block.
HDparm on Openfiler reads ~350MB read.
Test VM, win08 x64, 40GB vmdk
Guest disks tested thus far with IOmeter and HDtune. To keep things simple I've just been doing sequential read and/or write. VM on high gets ~45MB read (reviewing ESXi disk performance not VM test results) Openfiler iostat -d -x 5 3 shows 13% utilized during this test.
Now if I run multiple VMs with multiple HD tune tests the openfiler RAID and NICs become utilized and performance in my testing/reference VM remains ~40MB (not degrading much). My vmks will hit 70-75MB/s during multiple tests running. I can also run both HD tune and IOmeter 32k read at the same time in the test VM and almost double that VM's disk read to ~80MB/s.
Now why exactly doesn't a single HD tune and/or IO meter reach close to this level?
Could someone please explain to me what exactly ESXi is doing with the VM SCSI disk in this senario?
Thanks!
