4node : 2active 2-passive or a 3-active 1-passive setup,

[rehanshaikh]

Hi,

I’m new to clustering and would like some advise on how to move forward. The current setup we have is a 2 node active/passive setup with one virtual server called (exchange). To which the clients point to in outlook.


I’m having trouble deciding between a 4node: 2active 2-passive and a 4node: 3-active 1-passive setup, and how the virtual servers will work using the 3-active 1-passive or 2active 2-passive setup and then which virtual server will the client be needed to be pointed to.


Could someone please explain the difference between the 2 designs and which setup is best to use.




Thanks in advance

 

[xmsre]

1. With 2 actives and one passive, you are best suited for rolling udates. If your application allows it, in the event of failure of two nodes, both active instances will be running at a reduced capacity on the surviving node.

2. With two actives and two passives, you allow for failure of two nodes without reduced capacity/performance. You control what virtual servers can reside on which node through preferred owner and possible owner.

3. There is also the difference of the quorum type. For a traditional quorum (shared disk) the owner of the quorum wins arbitration of resource ownership. A traditional quorum is best suited when all the nodes are in the same physical location.

4. A majority node set quorum determines ownership of resources through agreement of a majority of the nodes. This is best suited for geographically dispersed clusters. When you do a GD cluster with a MNS quorum and an even number of nodes in each location, be aware of the failover behavior. If two nodes are on campus A and two nodes are on campus B, when one campus fails the cluster will never reach a majority. The cluster service on the suriving nodes will halt. This is by design. The MNS cluster has no way of knowing if this is a site failure or a communications failure. Halting the services on the surviving nodes when a majority cannot be reached allows for manual investigation and intervention. In fact, this behavior requires it. You have to manually start the cluster service on the surving nodes in this case. Normally, transferring to a DR site is something most organizations would like to have manual control over, so it does not present a huge issue. It's just something to be aware of.

 

[WhoKilledKenny]

Considerations for Active/Passive and active/active clusters for Exchange 2003.

Win2k Adv Server supports 1 EVS / 2 node cluster
Win2k Datacenter = 3 EVSs / 4 node cluster
Win2003 EE & DC = 7 EVSs / 8 node cluster

This is the trick. Exchange is limited to 4 storage groups per server. This is a physical limitation that applies to each node in a cluster. So for example, you would not create and active/active cluster if both nodes housed 4 storage groups each. You would never be able to fail over. In an Active/Active you coud have 2 storage groups on each or 1 storage group on one and 3 on the other, catch my drift? Here is an example of an active/passive with 3 EVSs on a 4 node cluster. Node names are clusternode1 thru 4. Exchange Virtual Serves are EVS1 thru EVS3 and are running on Clusternode1, 2, 3. Clusternode4 is passive. Each EVS is managing 4 storage groups. Let's say Clusternode3 fails, EVS3 would move to Clusternode4. At this point you need to get clusternode3 repaired as each remaining nodes has hit its 4 storage group limitation.
FYI:You can still use the fifth storage group, the recovery storage group for backup and recovery. It cant be used to create new mailboxes.

 

[WhoKilledKenny]

Regarding your clients. They will be pointing to the EVS on which thier mailbox was created.