If we install 2 multi-role Exchange servers in our building and a 3rd multi-role server in our remote data center, what is the best way to load balance them?  Do we need two load balancers or is there some way to span a single load balancer across the WAN ?

What about using Windows NLB as an alternative to using round robin internally?

Can a load balancer keep our interoffice Exchange CAS traffic from leaving our LAN and only failover to using the 3rd CAS/mailbox sever for internal users if both internal Exchange servers are offline?

We would also like remote users to "prefer" to use the data center CAS unless it is down. Right now we point our smart host directly to a CAS, but if we had a load balancer there, we could point the smart host to the IP of the load balancer and the load balancer could normally send it to data center CAS if it's up and forward it to one of the servers in the office otherwise.

Is it possible to do all this without a very complicated and expensive solution?

Depends... what is the connectivity speed between two sites, is it good enough?

You can use load balance in front of all the 3 CAS if your inter-site connectivity is very good.

What about using Windows NLB as an alternative to using round robin internally? WNLB and round robin is different,. You can use DNS Roud Robin if you want to or WNLB for all three CAS Server. Or Hardware loadbalancer for all three CAS servers

Can a load balancer keep our interoffice Exchange CAS traffic from leaving our LAN and only failover to using the 3rd CAS/mailbox sever for internal users if both internal Exchange servers are offline? If you want to use the load balancer then you don't need to fail them over one by one -- again you can use DNS Round Robing so the request will go to eah CAS servers one by one or use Hardware Load balance.

We would also like remote users to "prefer" to use the data center CAS unless it is down. Right now we point our smart host directly to a CAS, but if we had a load balancer there, we could point the smart host to the IP of the load balancer and the load balancer could normally send it to data center CAS if it's up and forward it to one of the servers in the office otherwise.

Use DNS Server and point the A record to the Primary Data center load balanced CAS server instead using IP or host file.

Hope that helps

http://blogs.technet.com/b/exchange/archive/2014/04/21/the-preferred-architecture.aspx

Talking about Best Practices: Set your office as primary site and remote DC as DR site.

Within you primary site, you need a hardware load balancer to distribute load to the 2 CAS. Windows NLB and DNS round robin won't work.

For remote DC, you don't need any LB since you only have 1 server.

Datacenter Switchover (failover to DR site) is a manual process.

The connection between the data center and office is around 50Mb, but it is also very heavily used and gets maxed out repeatedly. Actual file transfer speeds are much slower due to congestion.   We are looking for ways to reduce WAN traffic rather than upgrading the connection.  Reducing Exchange CAS traffic across the WAN and not adding a lot of replication traffic across the WAN to the data center when the DAG is implemented are options we are pursuing..

• Edited by MyGposts 5 hours 51 minutes ago

He specifically asked for using 3 CAS or not and yes he can- he needs to know what options are available based on this exis

Talking about Best Practices: Set your office as primary site and remote DC as DR site.

Within you primary site, you need a hardware load balancer to distribute load to the 2 CAS. Windows NLB and DNS round robin won't work.

For remote DC, you don't need any LB since you only have 1 server.

Datacenter Switchover (failover to DR site) is a manual process.

• Proposed as answer by Allen_WangJF 5 hours 33 minutes ago
• Marked as answer by MyGposts 5 hours 19 minutes ago

The connection between the data center and office is around 50Mb, but it is also very heavily used and gets maxed out repeatedly. Actual file transfer speeds are much slower due to congestion.   We are looking for ways to reduce WAN traffic rather than upgrading the connection.  Reducing Exchange CAS traffic across the WAN and not adding a lot of replication traffic across the WAN to the data center when the DAG is implemented are options we are pursuing..

• Edited by MyGposts Tuesday, February 17, 2015 5:57 AM