Slow Disks After Cluster Creation

[mgyton]

Hi,

Apologies if this isn't the correct place for this question, but I wasn't sure where else to put it....

I have a Windows Server 2008R2 SP1 cluster, consisting of two VMWare (ESXi 5.1) virtual servers, which is experiencing very slow disk performance (~10mb/s when copying to or from any of the cluster disks). If I destroy the cluster and bring the disks online on a standalone server I can get over 200mb/s. The disks are presented as RDMs, with the two nodes on separate ESX hosts and the disks in Physical Compatibility mode.

Could anyone explain why disk performance could become so poor as soon as the cluster is created? All the tests in the cluster validation report passed.

Not sure if this helps, but I have also noticed that when I turn maintenance on for each of the disks in Failover Cluster Manager, their performance improves dramatically.

Thanks

[mrdavidson]

I'm having the exact same problem.  Any ideas?

[weinstein5]

What type of storage are you using to store your virtual disks? How have you configured the storage? Did you follow the VMware Guide when setting up your cluster - http://pubs.vmware.com/vsphere-51/topic/com.vmware.ICbase/PDF/vsphere-esxi-vcenter-server-511-setup-...?

[memaad · Accepted solution]

Hi,

First go through best practice guide for MSCS by VMware. Just  quick check... if you are using Round Robin for RDM, then this could be performance issue.

Regards

Mohammed Emaad

[mgyton]

Hi all,

Apologies I never updated this with the solution - Mohammed is spot on, we had followed MSCS guide by VMware but we were using Round Robin for RDM which isn't supported for clustering it seems (although apparently it will be in a future release?!)

Many thanks,

Matt

[memaad]

Hi Matt,

Glad that I was able to pin point the issue. Yes, in vSphere 5.5, Round Robin is supported with MSCS cluster,

Refer this doc VMware KB: MSCS support enhancements in vSphere 5.5

Regards

MOhammed Emaad

[stormvm]

Thanks we had exactly the same issue. Disks performance approx 300 MB/s, then as soon as these disks were added to Windows 2012r2 cluster as RDM's performance dropped drastically to 10MB/s.  There is very little other information about this, apart from this thread. Changing path policy from Round robin to MRU ensured performance was back to approx 300MB/s