VI3 HA and DRS

[jpoling]

We are running VI3 with four hosts. Three of the hosts are running 3.0.2, one host is still at 3.0.1. Virtual Center is at 2.0.2. We currently are not taking advantage of HA and DRS, though we are licensed for it. I would like to take the steps to move to the next level and utilize DRS and HA. Reading through the Resource Management guide, the setup seems pretty straightforward. I do have a few questions:

1. Should I move forward with HA and DRS now or wait until 3.5 is released in order to take advantage of any new features and improvements?

2. Does DRS and/or HA require all hosts to be at the same version?

3. Setting up a cluster seems almost trivial. Can hosts be moved into clusters that are created without impacting running VMs?

4. Is there a recommendation for number of hosts in a cluster? i.e., with four hosts in our environment today is it best to create two 2-host clusters or one 4-host cluster?

Any insight is greatly appreciated.

Jeff

[FredPeterson · Accepted solution]

1. There is nothing outstandingly new from what we've seen that would cause you to delay implementing a HA/DRS Cluster. Just Do It!

2. Technically no, but your 3.0.1 VM's will complain about VM Tools not being up to date if they are on 3.0.2 hosts.

3. Yes, you can add to clusters while VM's are running. You cannot remove from a cluster unless the host is in maintenance mode.

4. There is a limit, but recommendation would come to down reasons. Do all hosts have identical storage, network and CPU's? With 4 hosts, just put them all in one cluster. In a two host cluster, the second host has to be able to take on all of the VM's from the other host in the event HA kicks in. You're adding 100% capacity to Host B if A goes down. But in a 4 host cluster, if A goes down, B C and D only add 33% each. While in your environment 100% addition might not be be a big deal, I'd rather not deal with it IMO.

[BryanMcC]

1. Go forward.. You can always upgrade at a later date

2. I currenlty run HA/DRS with three 3.0.1 and one 3.0.2... I think not

3. You can move them into the cluster but not out the host must be in maintenance mode to move out of the cluster.

4. Your choice.. I run 4 HA/DRS and 4 outside the cluster.. I find I get more use of my resources when the Hosts are not in the cluster due to buffer for VM migration in case of failing host. Depending on SLA I just spread my "mission critical" VMs accross the 4 hosts in the cluster and have the other 4 run my "less critical" VMs.

Help me help you by scoring points.

[virtualdud3]

1. If you have the licensing in place, I don't see any reason not to begin taking advantage DRS/HA immediately

2. I think a better solution is to simply upgrade the 3.0.1 host to ESX 3.0.2

3. Yes, hosts can be moved into clusters without impacting running VMs. If you want to remove a host from a cluster, the host must be placed into "Maintenance Mode", and therefore all of the VMs running on the host must be either power-off or VMotioned to another ESX server

4. There really aren't any hard and fast rules regarding the number of hosts in a cluster. It is a matter of available resources of the hosts in the cluster, and if you are using HA, the number of host failovers you wish to be able to "absorb". I would go with a single 4-host cluster, as you have many more options for the host you wish the VM to run on, and you will likely have additional fault-tolerance. Plus, management should be easier.

###############

If you found this information useful, please consider awarding points for "Correct" or "Helpful". Thanks!!!

[dwight]

1. Although 3.5 does provide a number of useful features HA and DRS work well in ESX 3.0.1 and 3.0.2. If you have the licenses I would do it now.

2. No, although you will see warnings about VMware tools. This also means if you can do a rolling upgrade to 3.5 later.

3. Yes, However, if you need to remove an ESX host from a cluster without removing the VMs from the ESX server, I.E. entering maintenance mode you need to disconnect the host from the virtual center before removing it at the virtual center. Otherwise it will complain the host is not in maintenance mode. About the only reason I can think of for removing a host from the cluster with VMs running would be if you want to migrate the host and VMs to another cluster and you want the VMs and Resource pools to move as well.

4. A bigger cluster is generally better. If you build 2 clusters of 2 hosts a failure of one host will cause a loss of 50% of the cluster's resources. If all 4 hosts where in one cluster a single host failure would cause a loss of 25% of the cluster's resources. That said there is a hard limit of the number of nodes in a cluster running DRS and HA (16 nodes). There are also hardware reasons (all nodes not connected to the same shared storage and networks) or you have multiple network security zones.

---

Dwight Hubbard (RHCE, VCP)