DRS&HA Clusters

[maia06]

Hi,

I have 4 ESX3 servers.

Is it possible to create one HA only cluster on server1 and server2 and then

create a DRS only cluster on server1 and server3?

[AndrewJarvis]

Sure - as long as you have the licenses, create 2 clusters, enable only HA on one, enable only DRS on the other cluster. Any reason you dont want both?

[hicksj]

Also, keep in mind the efficiencies you gain by having more than 2 hosts in a cluster when using HA...

With 2 hosts, you can only effectively use 50% of resources on each host.

With 3, it boosts to 66%, and with 4 you can run at 75% capacity.

If you have special requirements to keep particular guests from moving around, or don't want particular systems to restart during a HA event, you can impose restrictions on the process. This should be better (if it meets your needs) than setting up multiple cluster profiles.

[maia06]

Let me explain why I am interested in this subject :

I have two distant computer centers - A and B with ESX servers in each.

Suppose I have server1 and server2 in the center A, and server 3 and server4 in the center B.

I would like to have an HA cluster on all hosts, so that one of the center could fail over to the other. But I would like too keep my virtual machines in their computer center except in the case of a disaster. So I would like to create 2 DRS clusters - one on server1 et server2 in the center A, the other on server3 and server4 in the center B.

[hicksj]

Ah, you can't do that. Sounds like you're looking for something like "sub-clusters" with different DRS/HA profiles, OR, "cluster fail-over" - which isn't possible (yet).

Curious, do you have your storage replicated between the datacenters?

[maia06]

They share the same SAN datastores.

So, if I understand well, it is not possible to create sub-clusters - neither DRS sub-cluster in a DRS cluster, nor HA sub-cluster in a HA cluster.

It seems however possible to create one independent HA only cluster on, say, server 1 and server3, then create one DRS only cluster on server1 and server2.

[hicksj]

What is DRS providing for you here? Not really anything.

Sounds like you really want:

All servers in a HA cluster. Do not install DRS. Run ALL guests on server 1 & 2. If entire data center fails, they recover on 3 & 4 at alternate location.

(although, where is the shared SAN datastore?? If you lose your datacenter, you're toast if you're not replicating!)

I'm guessing above that one site is primary, the second backup, so you would prefer the processing to be localized at the primary. If you don't care where the VM's are running, then load up DRS also, and have all the VM's balanced across the hosts.

Again, you can't balance load across multiple clusters. So its all or nothing.

[maia06]

Sorry,

Every LUN on the SAN is mirrored in the alternate location.

I need DRS clusters to be able to work with ressource pools over multiple ESX servers and eventually for load balancing but only by computer center.

To avoid a misunderstanding let us say that the VMWare Infrastructure is distributed in two buildings in about 3 miles a distance between them.

[hicksj]

Ok. The problem here again is that load sharing is a Cluster operation, and so is high availability.

So, I would go with 2 clusters, one per data center. HA/DRS enabled on both. Then script or manually kick off the recovery process at your alternate datacenter, using the mirrored data, should your original datacenter go offline.

[maia06]

Thank you.