Poor Disk Performance on VMs; Host using PERC H310 RAID 5

I have a Dell R620 running 5 500GB 7.2K RPM SATA 3Gbps 2.5in Hot-plug Hard Drives in a PERC H310 controller (which I am beginning to regret buying).

I am getting awful read/write speeds on Linux VMs (Fedora 19, Cent 6 and Ubuntu 12.04) running ESXi 5.0 free license. Using Thin, Thick or Sparse VMDKs don't make a difference.

An example hdparm:

cached reads 9750 MB in 2.00 seconds = 4877.74 MB/sec
buffered reads: 42 MB in 3.44 seconds = 12.21 MB/sec

Also a dd if=/dev/zero of=test bs=1048576 count=2048 (to create an arbitrary 2GB file) took about a minute and a half!

I have a very similar host with a software RAID controller that has similar VMs which are much faster in disk speeds. I didn't set up the RAID, I only manage the server, but IT sets them up.

The Datastore shows an average 1.189 millisecond latency for reads and a 6.3 millisecond latency for writes. The datastore shows 1.89 TB so that is about correct for 5 x 500 GB drives in a RAID 5. Initially I was thinking maybe it isn't using all of the drives. What could be going on? I have asked IT to take a look at the RAID BIOS and see if there are some diagnostics we can run.

UPDATE I realized the one faster host has 10K RPM disks and the newer one has 7.2K

UPDATE 2 Is it possible that ESXi being run off of a SD card is causing performance to lag for disk operations within VMs?

Solution 1:

The Dell PERC H310 controller does not have the ability to use write cache.

H310 — Entry hardware RAID controller. Provides entry-level performance with no cache. RAID5 functionality with modest performance.

Your write performance will suffer as a result of this. This is a very bad controller for ESXi or virtualization using local disks.