1

I have one DFSR hub server with about 80 connections to 80 remote offices.

I want to add another hub server for load balancing and redundancy, but I don't want to split the connections between the two hub servers. I don't like the idea of having to maintain\administer two different blocks of connections across two servers.

edit Microsoft best practices say to divide DFSR connections between multiple servers (if there are many high volume connections as in a hub and spoke topology) because too many simultaneous connections to one server can cause performance issues. The only recommendation is to manually split the connections between servers. I would prefer to have multiple hub servers configured the same (all of the hubs configured with connections to all of the remote offices) and have something like the NLB service divide up the connections for me automatically.

What I want is to have 2 (or possibly more) hub servers, each configured with all of the 80 some odd connections, doing load balancing with something like the NLB service. I assume this would work because the DFSR hubs could dynamically split the connections between themselves and just replicate the updates received on either of their connections to the other hub.

I know you can do a failover DFSR cluster, but from what I have read, a Windows failover cluster does not do load balancing. So only one node in a DFSR failover cluster is actually active at a time- is this correct? If so it doesn't seem to meet my requirements for load balancing.

I have found no docs related to setting up a NLB cluster with the DFSR service. Is this even possible\supported? If so, can someone provide info regarding a best practices way a doing this or a doc?

If a NLB cluster is not possible is there a way to do like a DNS round robin with DNS SVR records or possibly some kind of "front end" load balancing setup.

Thanks a lot.

**edit I guess I was just looking for a set and forget setup. The single hub is handling everything fine for now, but we are planning on adding to our DFSR infrastructure.

I guess I'll have to decided whether to keep one hub and do a fail over cluster with a second, or break out the connections to a second hub.

I think I'll do a single hub with a fail over cluster for now. Then, if I decide to divide up the connections in the future, I could always add another fail over cluster (requiring 2 more physical servers) and divide the connections between 2 fail over clusters. More complexity than I wanted, but I guess that's the only way.

red888
  • 4,069
  • 16
  • 58
  • 104

2 Answers2

1

This cannot be done. Since the DFS-R topology is stored in Active Directory, there is no way to configure replication group members to use the NLB cluster name/address or any other type of load balancer for that matter. If you need a custom replication topology like you describe with a dual-hub for performance reasons, you will have to configure it manually. DFS at the scale that you are describing can become very complex and will need proactive management. There's no "set it and forget it" at that size, unfortunately.

MDMarra
  • 100,183
  • 32
  • 195
  • 326
  • Thanks for the helpful info. I updated my question to include the configuration I opted to go with. – red888 Sep 18 '13 at 14:20
0

Maybe I don't completely understand your use case here. But let me take a stab at it since nobody else has responded.

I'll assume you have 80 offices, each with a single replication set to talk to your hub. Because you can set the "sending member" in DFSR, you should be able to add in another server to supply the replicated data to the remote servers. It is what is referred to as "mutli-master replication"

So, if you had two "hub" DFSR servers (or any number for that matter), they would need to be in the replication group. Once that was setup, the "remote connections" servers would need to point to all the "hub" servers and would be able to replicate on demand as needed. They would simply sync to the first available server.

One thing you'd need to address is making sure the remotes don't try to sync data back to both "host hubs". Sites and Services would need to be configured correctly to address that.

Did I miss what you were asking?

MikeAWood
  • 2,566
  • 1
  • 12
  • 13
  • I'll edit my question to clarify why I actually want load balancing in the first place. This confuses me though: "...making sure the remotes don't try to sync data back to both "host hubs"". If all connections are bidirectional and the hubs are setup to replicate with each-other doesn't DFSR take care of everything? Why don't I want the remote site replicating to both hubs? Thanks. – red888 Sep 18 '13 at 02:23
  • 1
    you would rather the remote replicate to a single hub and the hubs replicated between each other.. otherwise, there might be a chance of having the remote push copies up to each of the hub servers doubling the amount of bandwidth needed/used. As far as your update, is your current server keeping up? Adding a second hub server would give you redundancy, but like MDMarra said, what you are asking for isn't a plug and play, but IMO, I think you might be overthinking it. You can tell half the servers to use one and the other half to use the other, but you'd have to fail them forward manually. – MikeAWood Sep 18 '13 at 08:15