On Fri, 2005-08-26 at 19:13 +0100, T. Horsnell wrote: > >On Fri, 2005-08-26 at 11:57 -0500, Rex Dieter wrote: > >> Craig White wrote: > >> > >> > FWIW - I didn't realize how much of a performance dog uw-imap was until > >> > I installed cyrus-imapd - and this was on my own mail server where I am > >> > the only user. > >> > >> What kinds of tasks were dog slow? Could it be because your mailboxes > >> were in the default compatible-but-horribly-slow mbox format? > >---- > >yes indeedy - default compatible-but-horribly-slow mbox format would be > >an accurate description. > > I'm interested in your experience with this. > Do you have any kind of statistics? How big is your mailbox, > how many messages in it, and how long did it take to display them? > > I know its all a bit subjective, but we have recently had > issues about mail-response and switching to cyrus-imap > was discussed. However, when I looked in more detail > at what was going on, I found many of our 700 users with > all their mail in the mail spool, others with Gigabyte inboxes > on their home disks and suchlike. When we addressed these > issues, mail response improved dramatically, even with the > compatible-but-horribly-slow mbox format (which we maintain > partly because of inertiia and partly because many users > want to read their mail with different mailers depending > where they are) ---- I always try to keep folders down to a max of 200-300 messages. I don't have any empirical statistics to offer, only anecdotal/perceptual experiences but on a mail server with one user and evolution 2.02 (and previous 1.4.5 with FC-2 I think it was), my INBOX with say 200 messages would take 10-20 seconds with uw-imapd (mbox). Cyrus changing to a mailbox is almost instantaneous. My attention didn't need to wander off for the few seconds each time I changed folders or even between emails in a folder. Also recognize that this type of non-scientific testing accounts only for the actual MUA that I am using and it's almost always Evolution. Other programs tend to cache different things and some may be faster. Forget inertia though for a second. With cyrus-imapd you also get... Mailboxes with ACL's - permitting you to share mailboxes (or subfolders or both) with other accounts - very granular ACL's too, by group/user/anyone/all/anonymous and a host of granularly controlled attributes... lrsipwcda (lookup, read, seen, insert, post, write, create, delete, administer) Public mailboxes (the super shared mailbox) quotas (not file system dependent quota system) automated options on account creation (subscribed folders, quota, sieve scripts) sieve vs procmail - procmail is probably more powerful but is too limited by the need to have a valid shell to be usable for a user and users cannot possibly administrate their own scripts. Sieve has web based user-level server side scripting in either web-sieve or ingo (part of horde) and it's phenomenal - users can actually set, start and stop their own vacation autoreply - no administrative intervention. Users are handled as virtual - no home directory needed, no shell needed. Self-contained. You can use valid users if you wish and could probably use MySQL or other - OR - you can use ldap users - but no need for a home directory or posix account. Hence - virtual users. I had a long series of procmail scripts that I had to recreate in sieve - no big deal and it's much simpler to maintain now. So - if you think that it's just because it kicks uw-imapd all over the place in terms of performance as the rationale for migration, that's really only one of the reasons. Craig
