topic Re: P4000, VMware and QoS disk I/O performance in StoreVirtual Storage

P4000, VMware and QoS disk I/O performance

Julius Verne — Fri, 28 Jan 2011 15:32:11 GMT

Hi, we recently setup a couple of P4000's with VMware. We're finding that if a VM really hits the SAN hard with a lot of disk activity that it negatively affects the performance of the VM itself and other VM's.

What options do we have regarding QoS of disk I/O? Does this require purchasing the Enterprise Plus version of VMware for storage I/O control? Is there anything that can be done with the SAN management utilities?

What would be a best practice with say having an Exchange server on the SAN--different LUN's for storage groups and then throttle based on LUN?

Thanks!

Re: P4000, VMware and QoS disk I/O performance

chadw_1 — Tue, 01 Feb 2011 23:11:04 GMT

I am having the exact same issue migrating files via the old san (via iSCSi initiator in Windows) to VDMK disks on the VSA. About every 30 minutes we get a "Maximum Latency Value Exceeded" message in the CMC, the Virtual machine "hangs" and then about 2 minutes later everything starts working again.

Re: P4000, VMware and QoS disk I/O performance

Julius Verne — Wed, 09 Feb 2011 20:20:43 GMT

Does anyone know how we can ensure one server accessing the SAN doesn't bring the storage repository to it's knees when an increased I/O load is present? Is there a way to throttle this? If strictly using the SAN with VMware, could this be achieved with VMware's Storage I/O control feature? Are there any other options? This is my only concern about the Lefthand solution. Thanks!

Re: P4000, VMware and QoS disk I/O performance

Jay Cardin — Wed, 09 Feb 2011 21:13:27 GMT

I am running into the same problem (high disk latency) when writing large amounts of data to the P4500, even with Storage IO enabled. It (SIOC) only helps if the LUN is a VMDK controlled by the same vsphere server. If you are using RDM or an iSCSI iniator inside a VM, it complains about a "non-VI workload" and can't do anything about it but set an alarm.

Other things to check:
1) Make sure that you are pointing to the VIP and have enabled the LUN for load balancing.
2) Make sure to setup ALB.
3) Enable Jumbo Frames and Flow Control if possible. If the switch can't do both, use Flow Control. (check the data path End-to-End to make sure that everything has Flow Control and Jumbo Frames enabled.)
4) Setup MPIO. Round Robin pathing may help.
5) Disk Alignment. Make sure your VM disks are aligned to the VMDK size. You don't want to make the SAN work harder than it needs to.
6) Make sure your switches aren't dropping packets because the port buffers are full.

Check out this blog for more info on SIOC: http://www.yellow-bricks.com/2011/01/20/enable-storage-io-control-on-all-datastores/