Preferred server with clustering Exchange

[beltmanjr]

Hi all,
I dont know much about clustering with exchange, but need to know if it is possible to have an active active cluster where 2 servers actually both work on the same exchange 'system' at the same time.

And if there is an option to tell the desktops which server to communicate with as preferred server? (i.e. one server is in the UK and the other in the USA I would like the UK desktops to communicate with the UK server for as long as it is up and running).

Thanks

 

[58sniper]

No. A cluster is designed to appear as one. Email clients are configured to connect to the cluster, not a specific node.

Additionally, you really can't have a stretched cluster like that. With 2003, they need to be local to each other.

If those servers already have Exchange on them, you can't make a cluster. Clustering is done at server setup. Not after the fact.

Pat Richard MVP

Plan for performance, and capacity takes care of itself. Plan for capacity, and suffer poor performance.

http://www.ucblogs.net/blogs/exchange/

 

[beltmanjr]

We have an opportunity to set up a completely new system.
I am looking at using VMWare and SANRAD to realize this.

So, the datastores will be identical at both locations and also appear identical to the servers.

But I am afraid that desktops in the UK will try to communicate with the server in the USA and the other way around, creating a very busy network.....

 

[58sniper]

Okay, but you'd still have to have the servers local to each other, so you still have all that traffic on the WAN links.

And Exchange isn't supported on VMWare.

Pat Richard MVP

Plan for performance, and capacity takes care of itself. Plan for capacity, and suffer poor performance.

http://www.ucblogs.net/blogs/exchange/

 

[beltmanjr]

Exchange is now supported on Infrastructure 3. Still I wonder if clustering is the best solution to this issue.

Perhaps it is better for each office to have their own exchange server with their own mailstore and only take over from the other one if it fails?

 

[58sniper]

If they are on separate subnets, you can't do clusters.

Except as described in this article, Microsoft does not test or support Microsoft software running together with non-Microsoft hardware virtualization software. For Microsoft customers who do not have a Premier-level support agreement, Microsoft will require that the issue to be reproduced independently from the non-Microsoft hardware virtualization software.

Right from the horses mouth, and I'm now aware of at least three separate instances where PSS terminated the call and told the customer to duplicate the problem in a physical environment and call back if the problem continues. And that wsa BEFORE doing any troubleshooting.

Pat Richard MVP
Plan for performance, and capacity takes care of itself. Plan for capacity, and suffer poor performance.

http://www.ucblogs.net/blogs/exchange/

 

[TechyMcSe2k]

I would recommend Exchange 2007. Much more flexible in what you are looking for.

Cliff, MCSE/MCSA/MCTS/CCNA/VCP/CCA

http://www.samtechconsulting.com

 

[xmsre]

Wow. Whatever you're doing, I want some. In MSCS, either one node or the other own a resource. In Exchange 2003 and prior, this means shared storage; In EXCHANGE 2007, this means application level replication. The difference is the amount of storage required (1x or 2x).


I think I'll head down to the Mellow Mushroom (It's a Georgia thing). It's time for an adult beverage...

 

[beltmanjr]

Just wanna get to a system that can't go down.
So servers that run in active-active mode with replicated shared storage would be very cool, but it is getting complicated since we actually have 4 sites and not one.

Though, reading a VMWare piece on Exchange it seems that for clients to connect to exchange via the internet doesnt cause too much traffic at all...

But it is long ago since I actively worked with Exchange and I have trouble remembering what options there are to get to a situation where my exchange server can go down and another would pick things up automatically without downtime.

 

[58sniper]

Just wanna get to a system that can't go down.

No such thing, really.

But it is long ago since I actively worked with Exchange and I have trouble remembering what options there are to get to a situation where my exchange server can go down and another would pick things up automatically without downtime.

Clustering can pretty much do that, but in 2003, the nodes have to be local to each other because of shared storage (not replicated). As I mentioned earlier, clients connect to the cluster, not a specific cluster node.


In 2007, you can use CCR, which gives you two nodes with dedicated storage. But the same principal applies - clients connect to the cluster. But with dedicated storage, you've removed another single point of failure. You can also use SCR, which lets you do log shipping to a remote server to provide for site resiliency. Same principal applies about clients, though. You can also do the same SCC method in 2007, just like 2003. However, in 2007, you can only have active-passive, not active-active.



Pat Richard MVP
Plan for performance, and capacity takes care of itself. Plan for capacity, and suffer poor performance.

http://www.ucblogs.net/blogs/exchange/

 

[beltmanjr]

Thanks for this all.
Just reading about merge replication and peer to peer replication on MSSQL2005. The answer to my needs, but I guess MS Exchange isn't prepared for this yet.

So, I guess clustering MS Exchange as described and having the clients connect to the cluster is the best idea. The traffic shouldn't go crazy... I hope.... and speed should be ok.

 

[58sniper]

With cached mode, there is the initial caching of the mailbox. But after that, traffic is negligible. Of course, you don't mention how many users there are, so.....

Pat Richard MVP
Plan for performance, and capacity takes care of itself. Plan for capacity, and suffer poor performance.

http://www.ucblogs.net/blogs/exchange/

 

[beltmanjr]

Ah, only about 150....