A design failure in Exchange causes the Exchange Enterprise Edition
not to be so scalable as intended. Microsoft is going to advise not to have more
then 1000 users connected to a node in an active/active cluster.
This was first mentioned in Win2k magazine's exchange newsletter by Jerry Cochran (Compaq Exchange 2000 expert and Win2k magazine author).
When I read that, I was just finished designing a 3000 users per node active/active Compaq Exchange cluster based on MS and Compaq docs, for a project I work for. I never saw anything on this limitation in the initial Docs. I tried to contact Jerry Cochran, but he didn't reply. So I posted questions in forums and someone contacted me and said it was true, he was in MS "non disclosure agreement" so couldn't say to much. It seems that there were concerns in fail-over when using more than 1000 users a node. He also said when SP1 is released the Docs will be updated with this new limit advisory. I was disappointed in the Scalability and "Enterprise"-ness of Exchange 2000 and was still hoping in some way it was a misunderstanding. I contacted Microsoft Engineers in Holland, they didn't know anything about this, but when they contacted US engineers I got the truth on the for Microsoft sensitive issue.
The problem arises on Exchange 2000 cluster. (Exchange 2000 Enterprise
Edition) The cluster-nodes run Exchange in a Virtual Machine (VM)
When MAPI clients (Outlook) connect to Exchange 2000 cluster box it fragments the memory in the VM, when a lot (more than 1000) MAPI clients do this, all memory may be claimed. So in an active/active cluster, when a node fails over to the other active node, the second instance (VM) of Exchange (STORE.EXE) may fail to start because there is no memory available.
In other words the issue is based on VM issues on a running cluster with Exchange, where there is a possibility that not enough VM will be available to tolerate a failover on an already active cluster node.
It seems this problem isn't simple to fix, it sits deep in the architectural design of Exchange or rather Store.exe. Therefore won't be fixed by a hotfix or any Service Pack. I guess MS needs to reprogram store.exe from scratch to overcome this.
You can understand this is a sensitive issue for Microsoft. They can't supply a fix so they must take down the specifications of their first AD Cluster killerapp.
That must have been a though decision.
With this limit fragmentation still occurs but with this user limit there is
an assurance that the VM will be available for an Exchange Instance to fail
This still is a good solution for High Availability sites with no more than 2000 users with good ROI because both machines are used.
With this limit fragmentation still occurs but when failover occurs the
Exchange Instance will start in a fresh VM with fresh clean Memory on the other
So for this method you're not limited by 1000 users but rather limited by the sizing of your hardware.
Manually restart STORE.EXE on the still working node, so memory will be freed
up and start unfragmented, then fail over the Exchange from the failing node.
Yeah right, so much for unattended High Availability. This means that the still working node is also interupted!
Use win2k Datacenter 4 node cluster using 3 active nodes en one passive.
Exchange 2000 on Datacenter is only supported with Exchange 2000 SP1.
Like I said NO FIX !!!
So that is why it won't be fixed until Exchange 2003 (codename "Kodiak")is released in 2002/2003. Kodiak will be based on SQL database engine instead of the current ESE.
When SP1 comes out the Exchange Whitepapers will be updated with the new limitations. Let me guess the title:
HOW TO INSTALL EXCHANGE 2000 NOT-SO-ENTERPRISE EDITION
Source: Microsoft and exchange forums
Deploying Exchange 2000 Clusters within Design Limitations
UPDATE: Microsoft SP1 cluster Whitepaper
Update ! MS SP2 cluster Whitepaper
Amsterdam, June 16 2001
Subscribe to Bink's Newsletter to keep you up to date!