Several important questions regarding SIOC / Storage IO Control

[Nitzan_S]

So I've read many articles, blog posts and even watched the VMworld 2010 session regarding SIOC, but I still have some important questions I would like to verify before I go ahead and implement this feature in my environments:

1. All the examples and use-cases I've seen discussed SIOC as a way to prioritize disk usage for critical VMs and such. I've seen only general remarks about regular same-priority VMs with a "noisy neighbor".
   So, to be more specific, in my environment I have lots of VMs which have the same priority and shares - no critical VM and such. We want to implement the SIOC in order to prevent one VM from hogging all the disk resources, impacting other VMs performance.
   Is SIOC efficient in this matter?
2. I understood that the IO control is being done altering the host array queue depth. Usually the max is 32, and SIOC cuts it down to adjust to the desired latency.
   My question is - if the queue depth really is host based, how well does SIOC work when there are lots of VMs hosted on the hosts (25-30 VMs per host)?
3. Last but not least, I understood that using SIOC on datastores that are shared across several vCenters is not officially supported but it is possible and usable by enabling the SIOC on all the vCenters with the same threshold value.
   Has any one tried it and can comment on how well it worked for him/her?

[mcowger]

1) Its well designed and supported for this use case.

2) Its per-VM queue not per host queue

3) Its not supported, and its really not a good idea.  Why would you want to do this?

[vPatrickS]

Hmm,

regarding to answer 2)

Isn't it based on the LUN queue/datastore?

Regards

Patrick

[depping]

vPatrickS wrote:

Hmm,

regarding to answer 2)

Isn't it based on the LUN queue/datastore?

Regards

Patrick

It actually is "datastore wide scheduling", meaning across multiple hosts.