Array controller failover (Dell md3020i, esxi5) and VAAI

[shane_presley]

We're running 6 esxi 5.1 hosts that use vmfs5 storage provided via iscsi from a dual-controller Dell md3020i.  Our load is about 900 IOPs, but can spike to 1800 IOPs.

During times of high load, our virtual disks failover from one controller to the other.  This doesn't create downtime, but does produce some problems about virtual disk balancing across the controllers.  We opened a case with Dell, and they see Ancient IO errors coming from the esxi hosts.  Dell is investigating a possible firmware upgrade (not yet released), and also suggesting a hard drive firmware upgrade.  They also mentioned we may be nearing the capacity of the array (in terms of IOPs).

We also opened a VMware support case.  They are recommending we disable VAAI. (http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=103366...).  However everything I've read talks about the performance benefits of VAAI, and our array officially supports it.  I ran that suggestion by Dell, and they disagree -- recommending we keep VAAI enabled.

So we're stuck with what to do next.  Any community suggestions, about VAAI or other changes to help with our array failover issue?

Thanks

[vmroyale]

Note: Discussion successfully moved from VMware vSphere Client for iPad to VMware vSphere™ Storage

[admin · Accepted solution]

I don't think VAAI is an issue. It might due to overloading on disk due to high IOs , it is moving to other controller. Please update the firmware..

Do you have log snippets pointing to ancient IOs which shows issues with ESXi

[shane_presley]

Dell confirmed this is a firmware bug (on the controller).  They don't have a fix out yet, but it's scheduled for August.