RHEV-H : Only half of CPU's available after hypervisor upgrade to v3.0.5

Latest response

Hello...me again.

 

I've just upgraded a hypervisor from v3.0.3 (20120320) to v3.0.5 (20120510) and have noticed that the available CPU's being reported by RHEV-M has been halved. Prior to upgrade I had 24 CPU's , now its showing just 12 CPU's.

 

The hardware itself is fine with both CPU's healthy and operation and with hyperthreading enabled.

The issue was highlighted when a VM with 16 vCPU's was unable to live migrate to the upgraded hypervisosor. RHEV-M complains that not enough CPU resource is available on the target hypervisor :-

 

Cannot migrate VM. There are no available running Hosts with enough cores in VM's cluster .

 

Anyone else seen this ?

 

Case :- 683080

Responses

Yes, RHEV-M in 3.0.5 will count only cpu cores available on a hypervisor without including hyperthreads.

 

The reason is that we cannot consider a hyperthread equal to a cpu core when we consider it for scheduling purpose by vcpus.

 

Later versions of RHEV-M may report cores and hyperthreads separatesly and allow over committing vcpus if desired. Still undere discussion.

Thanks Sadique

 

Bit of a backward step though!

 

My existing RHCS VM clusters allow me to allocate all hyperthreads available to a VM.

Moreover, where does that leave my VM's with 16 vCPU's ? 

 

They are now stuck on one hypervisor, unable to migrate to the upgraded node until I power them down, change their CPU allocation from 16 to 12.

I now have the unenviable task of asking my customers for downtime on their VM's whilst I remove 25% of their allocated processing bandwidth. (Wish me luck !!!)

 

It would be nice to know this up front as a pre-req to updating to 3.0.5. (20120510)

 

I see that my ticket has just been updated with the relevent bugzilla reports. I shall read......

 

Thanks again. 

Just to help out any future readers, a subsequent change was made to provide a configuration value for overriding this behaviour - allowing Hyperthreading to account for additional cores as it had in the past (pre-3.0.5).

Users wishing to learn about the pros and cons of doing this, and the mechanism for doing it, should refer to this knowledge base article:

https://access.redhat.com/knowledge/solutions/227953

The VDSM errata that included this latest change was:

https://rhn.redhat.com/errata/RHBA-2012-1390.html

Close

Welcome! Check out the Getting Started with Red Hat page for quick tours and guides for common tasks.