EDIT: This post has been edited on 5/19/2008 to add a note about pointing Exchange 2003 mailbox stores to Exchange 2007. Exchange 2007 saw a fundamental change in the elimination of a couple...
Published Mar 15, 2007
Version 1.0
Blog Post
Heidi,
It depends wholly on the physical layout of the boxes. By using an Ex2003 server as the “default” PFDB for user mailboxes, that specific Ex2003 server will perform the cost analysis for performing referrals. Note that a referral won’t happen if that same server also has a replica. Otherwise, that specific server perceives all the Ex2007 servers to be the same cost because they’re all in the same routing group. If indeed all the Ex2007 servers have the same cost then there’s nothing special you need to do. The trick here is to avoid a client referral off to never-never land when it can otherwise be prevented.
One technique would be to deploy (at least) a single Ex2007 PFDB server into each physical AD site. Then, after the hierarchy has replicated onto these new servers, adjust all the default PFDB settings for all mailbox DBs to point to one of the proximate Ex2007 PFDBs. Users shouldn’t notice much of any behavioral change as the Ex2007 server they’re now querying will a) always refer them somewhere else (as this is a hierarchy-only server) and the choice it makes for which server to send to will be consistently based on the AD site cost info, and not the legacy routing group cost info (which is how the Ex2003 server would behave).
The presence of Ex2003 servers in the org, or whether or not they contain replicas is essentially irrelevant. The only issue is which server the client asks for a referral, and then how that server computes referral costs. Since the Ex2003 servers are nearly guaranteed to produce wildly unusable referrals (given that all the Ex2007 servers appear to have the same cost), you’re better off upgrading, at the very least the PF hierarchy, to Ex2007.
You can then deploy further Ex2007 PF servers for content, and begin moving content from the Ex2003 servers at your leisure.
It depends wholly on the physical layout of the boxes. By using an Ex2003 server as the “default” PFDB for user mailboxes, that specific Ex2003 server will perform the cost analysis for performing referrals. Note that a referral won’t happen if that same server also has a replica. Otherwise, that specific server perceives all the Ex2007 servers to be the same cost because they’re all in the same routing group. If indeed all the Ex2007 servers have the same cost then there’s nothing special you need to do. The trick here is to avoid a client referral off to never-never land when it can otherwise be prevented.
One technique would be to deploy (at least) a single Ex2007 PFDB server into each physical AD site. Then, after the hierarchy has replicated onto these new servers, adjust all the default PFDB settings for all mailbox DBs to point to one of the proximate Ex2007 PFDBs. Users shouldn’t notice much of any behavioral change as the Ex2007 server they’re now querying will a) always refer them somewhere else (as this is a hierarchy-only server) and the choice it makes for which server to send to will be consistently based on the AD site cost info, and not the legacy routing group cost info (which is how the Ex2003 server would behave).
The presence of Ex2003 servers in the org, or whether or not they contain replicas is essentially irrelevant. The only issue is which server the client asks for a referral, and then how that server computes referral costs. Since the Ex2003 servers are nearly guaranteed to produce wildly unusable referrals (given that all the Ex2007 servers appear to have the same cost), you’re better off upgrading, at the very least the PF hierarchy, to Ex2007.
You can then deploy further Ex2007 PF servers for content, and begin moving content from the Ex2003 servers at your leisure.