 |
 | Home | Contact Us | FAQ | Search & Site Map | Link to Us |
 | Sign In | Join | Other 45 Sites in Network |
Windows Server Forum / Exchange Server / Design / January 2005Advice needed on design of fault-tolerant multi-site solution
Hi, I wonder if someone can provide any suggestions for the following design problem: I am planning a Notes to Exchange 2003 migration. The E2k3 servers will be new, dedicated, Active/Active/Passive clustered kit at each office connected to a SAN in each office. There are 2 geographically seperate offices, which are connected by a 34MB WAN link. The user community consists of 7,000 split evenly between the two offices. The requirement i'm trying to design for is to have all 7,000 mailboxes replicated at both sites (a Disaster Recovery requirement). So far as I can see the options are either: a) the cluster has to span both offices, which wont work because of the link speed b) each office has a seperate cluster, one being asynchronously replicated to the other (which wouldnt work because the Exchange services could only be provided from the 'live' cluster and so all 7,000 accounts would be accessing one office over the WAN) c) Could the Shadow Volume Copy Service in one office be used to keep a replica of the Exchange database up-to-date in the other? I can't really see a way of achieving this other than option C, anyone know if it'd work or can you see another alternative? Thanks for any answers Darren Darren, what is the actual requirement for availability? In other words, what exactly needs to be available when in the event of a disaster? Also, what do you define as a disaster (what would initiate this recovery process)? I ask because I find that people often talk about disaster recovery and want such things, but when it comes to paying for it, they find out they didn't really need quite that much availability. Thoughts inline: So far as I can see the options are either: a) the cluster has to span both offices, which wont work because of the link speed >>No, that's correct. With Microsoft clustering and Exchange you won't be >>able to span the cluster like that. I asked the questions above because >>depending on how you answer the question, you may need to get the data to >>the failover site which may mean upgraded link speeds. You can't beat >>physics, but you can design with them in mind. b) each office has a seperate cluster, one being asynchronously replicated to the other (which wouldnt work because the Exchange services could only be provided from the 'live' cluster and so all 7,000 accounts would be accessing one office over the WAN) >>What do you mean by that? This is often referred to as a geo-clustering >>scenario. Often, there is hardware involved and distance limitations >>because you would definitely need to get the data from geographical >>location to another within your required timelines. c) Could the Shadow Volume Copy Service in one office be used to keep a replica of the Exchange database up-to-date in the other? >> Up to date? No. Could it be part of your solution to provide what the >> requirements dictate? Yes. But you still have to be concerned with how >> you're going to get that data from site to site (see above pre, #1 & #2). If you can further refine your requirements, it would be helpful in providing thoughts on how to accomplish. As an example, if service is more important than legacy data, there are other options which might make this much easier. If you have to have legacy mailbox data (anything that was in the mailstore at the time of disaster) within minutes, then you'll need to work on the physics of transporting the data in a usable manner from geographical site to geographical site. -ajm > Hi, I wonder if someone can provide any suggestions for the following > design [quoted text clipped - 31 lines] > Thanks for any answers > Darren Do you mean that you're going to put a three node cluster in each office? You could get something like Veritas Volume Replicator and Veritas Clustering Services and Veritas Volume Manager all working together here, I know they have an Exchange solution. Mucho $$$ though I imagine (money isn't my department). I'm not sure how fast it would be over a 34 megabit link. Can you light up some dark fiber between the locatiosn and hook your san up tot hat? That would be easiest by far. Then we have 3500 users RPCing away over a 34megabit link, might break things a bit.  Signature--Brian Desmond Windows Server MVP desmondb@payton.cps.k12.il.us www.briandesmond.com > Hi, I wonder if someone can provide any suggestions for the following > design [quoted text clipped - 31 lines] > Thanks for any answers > Darren I just got off of a project that utilized Veritas VVR and Veritas VCS. It's a good product and works pretty well. The product is not that much more expensive than Windows Server 2003 Enterprise Edition (you can run your cluster on Windows Server 2003 Standard Edition just fine). I consider the need for training and for professional assistance to install it an almost necessity with Exchange 2003 clustering or Veritas clustering. It is hard to have a mission critical environment with untrained people...it's like going to perform heart surgery without a cardiologist. The downside is that it is an Active Passive (N+1) or Active Passive Passive (N+2), in your case, product. It will not run Active Active (N2 or N-squared) or Active Active Passive (N2+1 or N-squared + 1). Three suggestions: 1) Go to the training for Veritas VVR and VCS. 2) Hire out a Veritas consultant (usually a 1 to 2 week project...get 1 day of their time to work on the precursors (SAN, network, etc) a) MAKE SURE that the consultant is both Exchange and Veritas knowledgable. This is key. 3) Pay the support and maintenance fee (retail is 25% of the retail price) so that you have support and maintenance and upgrades. Bob > Do you mean that you're going to put a three node cluster in each office? > You could get something like Veritas Volume Replicator and Veritas [quoted text clipped - 42 lines] > > Thanks for any answers > > Darren Many thanks guys, that made for interesting reading. I've taken a look now at WANsync, Neverfail and GeoCluster but the problems as I see them are - WANsync/Neverfail - work on a server-to-server basis (just like an active/passive) but over distance. It doesn't look as though they'll cope with an MSCS source because of distance. I haven't looked at the Veritas stuff but I will defintiely look at that now. Geocluster looked promising but it seems to based on the 'stretched' cluster idea which i understood had distance limitations - the offices are 250 miles point-to-point so probably more that that from a physical transport point of view. From a Microsoft standpoint it looks as though MSCS is the weak link - it isn't geographically aware and so a software-stretching of the cluster isn't officially supported. That would bring me to the Microsoft HCL and something like EMC Symmetrix which does indeed look mighty expensive and I think would be out of the question. (can SRDF cope with distances of 500 miles or more does anyone know?) By the way, most products seem to use Async replication which Microsoft advocates against with Exchange. I can't really see a way of doing distance-based replication without it though unless you do implement a massively expensive hardware solution and try to do it synchronously with local buffering to avoid the i/o overheads. I've re-reviewed at the business case and the requirement is to have email provision the next day from the DR site for users, re-providing their mailboxes would be a secondary priority so the solution wouldn't need to be an instant failover.. Trouble is, I can't see a way of recovering mailboxes to a standby Exchange server if the original cluster has been destroyed. If there was a way, i'd think about perhaps doing a daily Volume Shadow Copy of the SAN onsite (for day-to-day quick recovery) and then sending a tape-copy down to the DR site daily in case of a disaster. I dont think I can have a seperate exchange server outside of the cluster yet part of the same Exchange organisation though can i? and even if i did, the restore is going to be pretty tricky - DNS aliasing to reroute clients, and modifying the mailbox servername in AD being the first two I can think of. Another option I can see is to do away with the cluster and build the resilience into the server - lots of CPUs, memory, redundant everything and load-balanced NICs to allow it to host the 7000 users. This single-server design should then allow the use of WANsync or similar. Al this is written without looking at the Veritas products which I will now go and do. It seems like a fairly major element of compromise might be needed here though :o( All the best, Darren > I just got off of a project that utilized Veritas VVR and Veritas VCS. It's > a good product and works pretty well. The product is not that much more [quoted text clipped - 72 lines] > > > Thanks for any answers > > > Darren |
Reply to this Message |
 Sign In
 Join
 My Latest Posts My Monitored Threads My Blog My Photo Gallery My Profile My Homepage Start New Thread Enable EMail Alerts Rate this Thread
|
©2009 Advenet LLC Privacy Policy - Terms of Use
This website includes both content owned or controlled by Advenet as well as content owned or controlled by third parties.