Exchange 2010 load balancing
Posted on 2010-11-18
We have very short time to make a decision about Exchange 2010 load balancing and HA.
I have been spending nights reading on Exchange 2010 as it is reletaviely new for me. I have some experience with Exchange 2007 but never with more than one exchange server.
Basically we have 4 locations to put exhcnage 2010 for. All these 4 locations are connected in a STAR topology using leased lines so they all can access each other, needless to say that dur to routing, they all have different internal netowrk addresses e.g. 192.168.1.x, 2.x, 3.x etc. You get the idea.
Now, I was planning to use 2 exchange servers with a hardware load balancer and implement DAG. But there, the hardware load balancer becomes a single point of failure. Not only that, if you think about it, there are 2 possible scenarios.
1. both MBX servers are in same site: SO the hardware load balancer and both mailbox servers are in the same site. So, the users at remaining 3 sites connect over the leased line to their mailboxes. So the leased line connection at the site with the exchange server becomes single point of faliure for the remaining 3 sites.
2. Put MBX server1 at say site 1 and then put MBX server2 at say site 2.: Here, we keep 2 exchange servers on 2 different sites. That resolves the leased line failure problem but, think abou thtis, even the users at site hosting the 2nd MBX server, will have to to the site with the hardware load balancer first which will then point them back to the mailbox server within their own site. Same thing for the other sites as well. The users who themserlves are at site 3 or 4, but has their mailboxes on the server MBX2, which is at site 2, they will also have to go to Site 1 for the load balancer and then from there, go to site 2 because the load balancer will point them there. Now, I know the loadbalancer can also point them to the MBX server in site 1, but not always.
So I came up with a different idea.
1. We setup 2 exchange servers with MBX, CAS and HUT roles.
2. Server1 is placed in site 1 and has database1, which has mailboxes for sites 1 and 3.
3. Server2 is placed in site 2 and has database2, which has mailboxes for sites 2 and 4.
4. The Outlook clients at sites 1 and 3 point to the Server1 and the clients at sites 2 and 4 point to server2.
5. The server1 has a DAG copy of the database2 and the server2 has a DAG copy of database1 and vice a versa. The replication betweent he 2 exchange server keeps happnening in the background.
6. Say the server1 failed for whatever reason. Hardware fault, software fault etc. Obviously, instantly, the Outlook clients in sites 1 and 3 will lose connectivity, because here, we don't have the load balancer.
7. My question here is that IF I was to divert the Outlook at sites 1 and 3 to point to server2, will they be able to connect to their mailboxes in DB1 which is the DAG copy on server2? I mean, do I need to do anything on the server 2 for this to work or because they already have the DAG copy of DB1.
8. This way, say for example, if the link at Site1 went down, we will still be able to re direct users at site 3 to connect to the server 2 at site 2, while we fix the link.
All servers are part of one AD domain, there are no domain trees or subdomains or anything at all.
I am open to discussion here, but I need to understand if tihs will work.
I know manual switchover is a pain but I have some plans for that.
My main reason for this approach over hardware load balancer is that link is not a single point of failure here as each site also has Internet connection which we can make user of temporarily. Also, a particular piece of hardware also can't be a signle point of failure. Not only that, we also divide the load on the servers.
So basically, this is a perfact design but I am sure, I am dreaming here and someone will bring be back right down to earth.