Re: Active/Active clustering?
- From: "Bharat Suneja [MVP]" <bharat@xxxxxxxxxx>
- Date: Wed, 24 Oct 2007 18:03:16 -0700
You have to understand two things here:
- How clusters work in terms of ownership of resources, including the quorum. Windows Server 2003 supports 1) (Traditional) Shared quorum 2) Majority Node Set (MNS) and 3) MNS with File Share Witness (with hotfix KB 921181 or SP2).
- Active/Active is only supported on 2 nodes, making quorum option #2 (MNS) irrelevant. If you add a 3rd node, it's not A/A any more.
- Not quite sure if option 3 has been tested with Exchange 2003 or if it even works - conceptually, it should.
- Regardless, if you use a shared quorum or a FSW, it resides in one data center.
- If and when the link between the 2 snaps, NODE1 in BLDG1 will own the quorum. The shared storage in BLDG2 used by EVS2 on NODE2 will not be accessible to NODE1 (believe that'll be over the fibre link as well). EVS2 will not fail over to NODE1.
- NODE2 in BLDG2 will not be able to stake claim over the resources (EVS2 - which it hosts) because it doesn't have quorum.
Other than that, Active/Active isn't recommended even by Microsoft - it suffers from scalability and virtual memory fragmentation problems.
Would recommend looking at Exchange Server 2007 - the built-in replication functionality will help you achieve higher availability, with or without clustering.
Alternatively, look at 3rd party solution that use synchronous & asynchronous replication to provide such functionality - with or without clustering.
--
Bharat Suneja
MVP - Exchange
www.zenprise.com
NEW blog location:
exchangepedia.com/blog
----------------------------------------------
"Phil McNeill" <philmcneill@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message news:%23Ar74flFIHA.5980@xxxxxxxxxxxxxxxxxxxxxxx
So, I am reading some bad things about this kind of setup. Our network configuration is such that we have two buildings with a single fibre link between them. We have domain controllers at either end to service login requests for the local users, and I thought it would make sense in a two node cluster to have one in each building, and the database split so during normal operations each server services clients in the same building. In my simplistic view of things, this seems to accomplish both high-availability and disaster recovery in the event that one server, or one whole datacenter goes away.
Since there is only one network link between the buildings, both public network and heartbeat travel across the same physical link. So, in the event of a failure of that link, what happens to my mail? I am assuming in an Active/Passive setup, the active server continues as normal, with only the redundancy lost. Clients sitting in the passive node's building lose access to mail (since it's all running in the node at the other end of a broken piece of fibre). Please correct me if I'm wrong there.
In an Active/Active setup, would all clients retain access to their local database? This is the big advantage to going Active/Active that I can see.
Should I give up on the idea of geoclustering altogether and just keep the two boxes beside each other in the same datacenter and come up with a different DR solution (standby server at remote location?)
Opinions/tips appreciated.
Thanks,
Phil
.
- Follow-Ups:
- Re: Active/Active clustering?
- From: Phil McNeill
- Re: Active/Active clustering?
- References:
- Active/Active clustering?
- From: Phil McNeill
- Active/Active clustering?
- Prev by Date: Re: Shrinking Inbox and Keep old mail available
- Next by Date: Re: hot to "turn off" mailbox.
- Previous by thread: Re: Active/Active clustering?
- Next by thread: Re: Active/Active clustering?
- Index(es):
Relevant Pages
|
