new AMD CPUs and EVC woes. What now?

[taylorb]

Here's where I am right now.

• 7 hosts with EVC in Opteron Generation 3 mode added over the past 3-4 years using Istanbul and Magny-Cours based processors.
• 150 running VMs, many business critical
• 3 new host with new Bulldozer processors,

I've tried putting the new hosts in the exisiting cluster and it won't work.  Apparently the new Bulldozer chips do not support the Generation 3 EVC mode, but a lower subset called Generation 3 with no 3D now.  I tried making a new cluster with Gen 4 EVC, and they don't Vmotion.  I tried dropping it to Gen3 (no 3Dnow) and they won't vmotion from the regular Gen 3 hosts.  I presumably can't drop to EVC mode Gen3(no 3dnow) on the existing hosts while they are all up and running with 3dnow enabled. 

So it would appear I need to somehow drop the existing VMs to the Gen3(no 3Dnow) mode as that is the latest one that supports all of these processors.  How do I get there from here?   I mean, I can't shut down all 150 Vms and put the cluster in a new EvC mode. What would you guys recommend?  I don't know whether to blame AMD or Vmware, but EVC is supposed to eliminate these kinds of problems, right?

[sparrowangelste]

you could create a new cluster with the new EVC and the new hosts.

then during downtime or scedualed outage move over powered off vms slowly to the new cluster.

or

down all the vms and then change the evc.

[markdjones82]

You should be able to change the EVC mode down on the cluster as long as the hosts support it, but the guests may not migrate until they are rebooted with that new instruction set.

[taylorb]

So you think it will let me switch from the Gen3  to the Gen3 (no 3Dnow) at the cluster level and then each VM will switch to that processor mask at their next reboot?  All CPUs I have shoudl support that no 3Dnow mode. 

It gets a little hairy clicking buttons on production stuff to see what happens!

[taylorb]

Looks like that isn't going to work.  I guess I am going to have to put the hosts in a new cluster with the Gen3(No3Dnow) enabled, cold migrate the VMs until I vacate a host, and then move the host into the new cluster with the lower EVC.   Then repeat for probably the next 6 months one host at a time.

Not thrilled about this.

[meistermn]

1.) So do you want to move from Istanbul and Magny-Cours to the new piledrivers (8 or 16 cores) ?

2.) If you create a new cluster do vmotion from vsphere 5.0 8old cluster) to vsphere 5.1a ?

[meistermn]

Found following hint in KB

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=100321...

For AMD processors, there are two different sequences of EVC levels: Levels A0 to A3, which support the first 3 generations of AMD Opteron™ processors, and levels B0 and above, which support AMD Opteron™ Generation 3 and later processors. Note that the AMD Opteron™ Generation 3 (no 3DNow!™) EVC baseline is common to both sequences. For example, an AMD processor that supports the AMD Opteron™ Generation 4 EVC baseline can support the AMD Opteron™ Generation 3 (no 3DNow!™) baseline, but not the AMD Opteron™ Generation 3 baseline. An AMD processor that supports the AMD Opteron™ Generation 3 baseline can support the AMD Opteron™ Generation 3 (no 3DNow!™) baseline, but not the AMD Opteron™ Generation 4 baseline. AMD EVC baselines are listed in Table 1.2.

Also table 2.1 is interessting:

Table 2.1: AMD EVC Baselines supported in vCenter Server releases

	EVC Cluster Baseline
vCenter Server Release	AMD Opteron™ Gen. 1	AMD Opteron™ Gen. 2	AMD Opteron™ Gen. 3	AMD Opteron™ Gen. 3 (no 3DNow!™)	AMD Opteron™ Gen. 4	AMD Opteron "Piledriver" Gen.
VirtualCenter 2.5 U2 and later updates	Yes	No	No	No	No	No
vCenter Server 4.0	Yes	Yes	Yes	No	No	No
vCenter Server 4.0 U1 and later updates	Yes	Yes	Yes	No	No	No
vCenter Server 4.1	Yes	Yes	Yes	Yes	No	No
vCenter Server 5.0	Yes	Yes	Yes	Yes	Yes	No
vCenter Server 5.1	Yes	Yes	Yes	Yes	Yes	Yes

[taylorb]

meistermn wrote:

1.) So do you want to move from Istanbul and Magny-Cours to the new piledrivers (8 or 16 cores) ?

2.) If you create a new cluster do vmotion from vsphere 5.0 8old cluster) to vsphere 5.1a ?

1) I think they are Bulldozer chips, but yes

2) It fails because Gen4 does not support ALL the features of Gen3.  If I had somehow predicted the future and selected Gen3(no 3Dnow) back in 2009 then I would be OK.

Basically AMD screwed us over by removing some features from the newest chips that had been present in the old ones, which has not been the case in the past.