EVA 4400 / Windows 2008 Cluster / MPIO using HP DSM

[Matthew, Millers]

Hi,

We have a EVA4400 access by a 2 node windows 2008 cluster.

We have 48disk / 1 storage group / 1 vdisk (RAID10). The I/O profile is mixed between random (50%) and sequential IO (50%).

I believe that the deployment has been made without any extra configuration on any subsystems.

Workload is SQL server 2008.

We are noticing spikes of 3000 write operations/sec, what is concerning is the read latency spiking to 160ms. I would have thought in the RAID configuration above that we would be able to sustain the write operations with less latency.

I was wondering if anyone could advise on general configuration settings on the storage/HBA/switch/DSM to alleviate what appears to be a performance bottleneck.

I appreciate any advise people may wish to give!

Thanks
Matthew

[Víctor Cespón]

A few questions first:

- What kind of drives? in VRAID 1 you can expect 115 Iops per drive with 15k FC drives, and half that when using FATA drives
- Only one LUN? Many HBA drivers and operating systems have a limited queue per LUN, so all I/Os go to a single queue of 16 or 64 for example.
- Even with 48 disks, 3000 Iops mean 125 Iops per drive in VRAID 1
- MPIO set to load balance on all paths?

[IBaltay]

Hi,
here is the bunch of sizing tools for your inspiration:
http://h71028.www7.hp.com/enterprise/cache/576785-0-0-225-121.html?jumpid=reg_R1002_USEN

[Matthew, Millers]

Hi Guys,

Thanks very much.

It is possible that the queueing for the single LUN is causing the problem. Although it looks like we may just be topping out the IOP capacity on the diskset.

To degate any queueing issues, we are going to split into seperate LUNs and spread the datagroups across those.

Thanks again for your assistance, I will update and let you know how we are coming along.



Regards
Matthew