Now that Exchange Server 2010 has had its first birthday, it's a good time to remind folks about the built-in features for high availability, site resilience and disaster recovery in Exchange 201...
Published Dec 16, 2010
Version 1.0
Blog Post
Installed Exchange 2010 SP1 Rollup 2 on one of the CAS servers in the CAS Array yesterday expecting the clients (OL2007) to be unaffected, but in fact they stayed disconnected for the time the services were down being patched - about 5 minutes. Clients are in NON cached mode. This caused a major problem and lack of trust in the product.
I am suspecting that the CAS Array is set up fine, as is the client referral (checked all MBX DBs have the correct CAS Array assignment for RPCCLientAccess), and that the problem here is the way Windows NLB works. It is set up as recommended with source IP affinity for RPC connections, but what I suspect happens is that because the server still responds - i.e. it is only the SERVICES which are down and NOT the server, the client affinity persists and so the client shows as disconnected - even though it is still connected to that server in the CAS Array - does that sound reasonable ?
If this IS the case, is there any workround other than reconfiguring the whole array to move the server being patched out (which I suspect will require downtime anyway) ?
I am going to test this theory next week by seeing if the same clients have the same problem if I physically down the server in the CAS Array - in which case NLB should reconfigure to point to another server, so maybe 30 second interruption maximum ?
Anyone else experienced this ?
I am suspecting that the CAS Array is set up fine, as is the client referral (checked all MBX DBs have the correct CAS Array assignment for RPCCLientAccess), and that the problem here is the way Windows NLB works. It is set up as recommended with source IP affinity for RPC connections, but what I suspect happens is that because the server still responds - i.e. it is only the SERVICES which are down and NOT the server, the client affinity persists and so the client shows as disconnected - even though it is still connected to that server in the CAS Array - does that sound reasonable ?
If this IS the case, is there any workround other than reconfiguring the whole array to move the server being patched out (which I suspect will require downtime anyway) ?
I am going to test this theory next week by seeing if the same clients have the same problem if I physically down the server in the CAS Array - in which case NLB should reconfigure to point to another server, so maybe 30 second interruption maximum ?
Anyone else experienced this ?