On Mon, 29 Dec 2003, Jim Lawson wrote: > We're attempting to replace our existing mail server here with a cluster > of Linux nodes running sendmail/procmail/UW IMAPd. The Linux nodes use > are attached to shared storage (SAN) and using Sistina's GFS to manage > the locking.
I have never seen any sort of network filesystem clustering work out well for IMAP servers. Nor have I seen network locking work out well. mbx format is probably not your answer; it does shared write which is almost certainly guaranteed to lose on a network filesystem. Let's put it this way; if your vendor says that it will work, make them put that claim in writing before you do it. You might get some free hardware out of them before they throw in the towel. In general, it works better to split the mail store across a set of machines, each of which has direct access to its own piece of the mail store. At UW, we use the DNS to direct users to the right server; mrc.deskmail.washington.edu is the IMAP server which has mrc's mail, the machine may change as mrc gets moved about, but the name that mrc uses to get to his mail is always the same. Remember that IMAP servers are generally not CPU bound; they are I/O bound. Thus, fast access to the filesystem is more important than CPU load balancing. Our IMAP servers tend to have low load averages but very high disk/network utilization. -- Mark -- http://staff.washington.edu/mrc Science does not emerge from voting, party politics, or public debate. Si vis pacem, para bellum.
