We're consolidating our mailboxes from remote sites onto one Exchange Server 2003 system at a central data center. Should we make changes to our Active Directory (AD) topology as well?
Not necessarily. When Exchange 2000 Server shipped, Microsoft's recommendations were simple: Put at least one Global Catalog (GC) server in every AD domain that hosts Exchange servers or clients, and allocate one processor of GC capability for every four equivalent processors of Exchange capability. For example, if you have four dual-Xeon 2GHz Exchange servers, you should have the equivalent of two 2GHz CPUs for GC use. (The original recommendation was actually one GC for every four Exchange servers, but experience showed that the demand Exchange makes on a GC required a different rule of thumb.) Many deployments extend these recommendations and put GCs in the same AD site as clients to take advantage of the way that Exchange looks for GC servers. The server-selection algorithm always prefers GCs and DCs that are in the same AD site as the Exchange server; Microsoft Office Outlook 2003 shows a similar preference on the client side. But as long as you meet the minimum recommendations, you needn't make any other changes.
The whole point behind site consolidation, though, is to move as many GCs, DCs, and Exchange servers as possible to a central location. In consolidated deployments, having the GC servers and DCs in the same AD site as the Exchange servers becomes more important than in nonconsolidated deployments because the consolidated servers will generate a greater load. Therefore, if you're moving your Exchange servers to a central location, it makes sense to keep GC servers and DCs at remote sites. Depending on the number of users at those sites, you might want to keep DCs there to handle logon traffic. But you don't need to keep Exchange servers at the remote sites, as long as you have enough bandwidth to provide good client performance. And if you're using Outlook 2003, keep in mind that its Cached Exchange mode does a great deal to smooth out fluctuations in network bandwidth and latency.
—Paul Robichaux
End of Article
Register
