On Thu, 7 Dec 2006, Chris Miller wrote:
Using imapd-2006d (376) and pine with the new c-client has helped a bit. It is still noticably slower than imapd-2004g, but in a nearly-tolerable way.

OK, thanks for confirming that imap-2006d remedies the problem.

In some cases, procmail may be delivering directly to one or more of my problem mailboxes; would that have a similar impact?

Yes, it will. Anything that appends to a mailbox without assigning a UID forces a subsequent UIDPLUS-capable append to assign UIDs to those UID-unassigned messages before it can assign a UID to the message it wants to append.

   If you don't mind explaining, what is going on that causes the slowdown?

(1) Traditional UNIX format was never intended to scale to GB size mailboxes, and it doesn't. Some people will claim that flat files don't scale to GB size mailboxes.

(2) UIDs must be strictly ascending in the mailbox. A UID can not be assigned for a newly-appended message until all previous messages have an assigned UID.

(3) Appending to a mailbox without assigning a UID screws up the requirements of (2), and something has to go and fix it.

Do the UIDPLUS ids have to be re-evaluated for the whole mailbox if a message is found without one?

The simple answer is "yes". The more complicated answer is that messages without UIDs have to have them assigned, which entails some amount of mailbox rewrite.

> There is a smaller hit with traditional UNIX format that is unavoidable. It > should be much less costly than the big hit; however, if you really have > 940MB traditional UNIX mailbox format mailboxes you should seriously consider > using another format for other reasons. In my personal case, these are my exception mailboxes; mostly write-only, and cleared out a couple times a year. However, I know that I have some users who have primary mailboxes of this size and worse.

That should be motivation to get them to switch to a better format! :-) For what it's worth, the flat file mbx format will work much better if they don't want to go all the way to mix.

However, in general, flat file mailboxes should not be allowed to become that large.

I will say that the mix-format mailboxes I've been testing are working quite well, and very fast, aside from one situation where copying 1000 messages from a unix-format mailbox to a mix mailbox w/ pine took 25 minutes. I'm assuming that was a fluke, and will ignore it. :) As for the mix format more generally, are you confident enough in it to advise me to recommend it to my users when they complain about their mailboxes being slow?

Yes.

However, note that neither mbx and mix can be used over NFS. That may be an issue.

-- Mark --

http://staff.washington.edu/mrc
Science does not emerge from voting, party politics, or public debate.
Si vis pacem, para bellum.
_______________________________________________
Imap-uw mailing list
[email protected]
https://mailman1.u.washington.edu/mailman/listinfo/imap-uw

Reply via email to