The LISA 2000 invited talk "Design and Implementation of Highly Scalable E-Mail Systems" gave the theory of building a scalable IMAP implementation, but there wasn't any practice to go along with that theory. The purpose of this talk is to bring in some real-world experience.

The three major open-source IMAP servers for *nix (UW IMAP, Courier-IMAP, and Cyrus) will be discussed. The focus will be on scalability, and comparing and contrasting with some commercial alternatives. Two examples of the DIHSES architecture will be presented, as well as a similar architecture as modified for an Enterprise customer in the Netherlands.

However, there are also hidden/unexpected costs and requirements that go along with implementing a customer-visible IMAP service. Things that may work for an Enterprise or a commercial outsourcing firm do not necessarily scale well in an ISP or "free service" model.

Just because something is technically possible doesn't necessarily mean that it is feasible to do so with adequate backups and failover, or that it can be done without raising excessive liability or privacy issues.

Just because we can do something, doesn't mean we should.

AOL and Gmail will be used as contrasting models -- of what not to do.

Brad has specialized in Internet e-mail and DNS for many years, and has provided the benefit of his experience to the US Department of Defense, America Online, and Collective Technologies, among many others.

He has spoken at SANE'98, LISA 2000, UKUUG Winter Conference 2001, LISA 2002, BSDCon Europe 2002, and RIPE 44; was on the program committee for SANE 2000 and SANE 2002; and was a reviewer for the books _sendmail_ and _DNS and BIND_ (both published by O'Reilly).

He is currently setting up his own consulting company and working on writing his first book, among many other things.