IIRC Active/Active was the initial recommendation under E2k, until some of these MAPI session limits starts cropping up, and with memory fragmentation in ESE. We were in discussions with MS about deploying a large Active/Active cluster during this period, and they changed their tune part-way through the discussions and recommended Active/Passive rather than Active/Active.
With all of the redundancy we were building in with the Exchange design anyway (see below), clustering was only going to save us from physical hardware failure of one of the front-end servers, but we pretty much had that covered anyway. As others have said, clustering doesn't protect you from DB corruption. Config for single mailbox server was: Multi-Processor Server ECC Memory Multiple hot-plug power supplies on different physical power curcits, running on building UPS' (optionally local UPS for each power supply as well) Multiple hot-plug fans for cooling RAID'ed disk (obviously), on multiple channels, with hot spares Multiple Physical, Multi-Port NICs using port aggregation over multiple redundant switches. We also had a hot-standby server incase of total-systems failure (move the entire disk array to new server), and hot spares of every major component. We also deployed multiple Exchange servers, so in the case of complete failure of one server, only portions of the organisation were affected. We ensured that members of the same work area were located on physically seperate machines, so if one server did fail, at least one mailbox in each section was still able to send and receive important emails. Multiple incoming and outgoing connector servers provded some measure of protection against a single connector server failure (the fault-tolerance levels of these machines were much lower to save costs). All Public folders were replicated to at least two different servers. Even with clustering, a number of the measures described above were still required, such as building some level of redundancy into the front-end servers, but you now have the added complexity of the redundancy and fault tolerance required for the SAN device. My gripe with clustering is that there is a tendency to try and throw the entire org onto a single cluster (either Active/Active or Active/Passive), but when the cluster itself fails, the entire organisation is off the air. Something that isn't tolerated these days with mission critical mail systems. Exchange has so many built-in redundany and load sharing features, that clustering just introduces unnecessary complexity into the mix, especially in recovery scenarios. That being said, when I have deployed them and got the kinks worked out, they have been pretty solid. 2k3 may be a different kettle of fish. My $0.02 G. ----- Original Message ----- From: "Martin Blackstone" <[EMAIL PROTECTED]> To: "Exchange Discussions" <[EMAIL PROTECTED]> Sent: Saturday, June 28, 2003 11:34 AM Subject: RE: Clustering... is it worth it? I believe they have always recommended an Active/Active cluster. Paul Roubicheux sais the E2K3 clusters awesomely. -----Original Message----- From: Schneider, Bryan D [mailto:[EMAIL PROTECTED] Sent: Friday, June 27, 2003 6:14 PM To: Exchange Discussions Subject: RE: Clustering... is it worth it? You have the benefit of quick recovery in event of hardware failure on the server (not likely typically). But, it is really nice for maintenance where you have to apply patches, security updates, virus engine updates, service packs, etc... You can failover in a matter of seconds and you have as much time as you need to work on the server without interrupting users or bouncing email. On an active/active cluster we host 16,000 users, 2500 using Outlook and the rest using OWA 2000. We can have both virtual machines running on one quad-Xeon 700Mhz without users noticing much of a slowdown at all. Exchange 2003 with Windows 2003 runs more efficiently so far in our tests. However, Microsoft is now recommending ACTIVE / PASSIVE so you have a fresh server to failover to. You already have a key component - SAN - so I would cluster in a heartbeat. We haven't had any issues - except for a corrupted db which we attributed to the SAN. 2003 promisses to make clustering better, but we haven't tested that yet. -----Original Message----- From: William Lefkovics [mailto:[EMAIL PROTECTED] Sent: Fri 6/27/2003 7:10 PM To: Exchange Discussions Cc: Subject: Re: Clustering... is it worth it? But do consider revisiting this with 2003. With Microsoft running 16,000 users on an 8-node cluster now. Windows2003 and Exchange2003 of course. ----- Original Message ----- From: "Martin Blackstone" <[EMAIL PROTECTED]> To: "Exchange Discussions" <[EMAIL PROTECTED]> Sent: Friday, June 27, 2003 5:04 PM Subject: RE: Clustering... is it worth it? > That's pretty much the argument against clustering. > In fact, many folks will tell you that Exchange needs much more hand holding > in a cluster. > > -----Original Message----- > From: MSX dude [mailto:[EMAIL PROTECTED] > Sent: Friday, June 27, 2003 4:50 PM > To: Exchange Discussions > Subject: Clustering... is it worth it? > > Upper management here is inquiring about clustering our exchange server. > We already have our PRIV, PUB and DIR on a SAN. I don't see the benefit. > If the server itself fails, I can rebuild it in an hour. If the database > corrupts it would have taken the cluster down too. > > I have searched the internet but all I find are vendors praising clustering > because they want to see you something. Does anyone have any links or > whitepapers are unbiased in their opinions? > _________________________________________________________________ List posting FAQ: http://www.swinc.com/resource/exch_faq.htm Web Interface: http://intm-dl.sparklist.com/cgi-bin/lyris.pl?enter=exchange&text_mode=&lang=english To unsubscribe: mailto:[EMAIL PROTECTED] Exchange List admin: [EMAIL PROTECTED]
