iSCSI performance on Hyper-V cluster still not great?

[JonMoore87]

Hello!

I'm having some problems with I/O performance on our hyper-v cluster.

We have 3 hosts each running 3 guests connected to our DSS v6.
Each host has a direct 1Gbps connection (crossover cable) to a NIC on the DSS, the DSS has 4GB ram (although on the open-e status page it doesn't look like this is all being used) and 16 x 1TB SATA drives connected to an LSI 3Gbps RAID card in a RAID 6. The VG is a single LUN Block I/O

I'm when I look at my guest servers in perfmon i'm getting around 0.250 sec/read (and write) I understand that anything higher than 0.020 is regarded as poor performance.

The guest servers do feel pretty slow, which is what lead me to investigate.

Would switching the DSS to a single 10Gbps and putting in a 10Gbps switch for the hosts be the only thing that could solve this?

Or can anyone offer and tuning or config advice that will avoide the huge investment of 10Gbps network equipment for me?

Many thanks for all of your help,
Jonathon

P.S more info available on request... just let me know what you need (although it's a live system so I can't interupt services)

[enealDC]

Before you get into testing guest performance, you really should test storage performance from the perspective of the Hyper-V servers.
That should be easy to do..

[dprescutti]

Have you looked into Multipath I/O with multiple 1 GB NICs per host? There's good documentation here:

http://www.open-e.com/solutions/open...-windows-2008/

[webguyz]

Do you have Jumbo Frames enabled at both server and DSS nic cards?

Download IOMETER and run it from the host server and then from the guests. Google iometer to get tips on how to set it up.

When running a data copy from your host, take a look at task manager -> networks and see if your iscsi NIC cards utilization is 50% or better.

Good Luck!