IMAP debate/thread-that-wouldn't-die

[ken at flyingtoasters.net]

Ben said:

>   All I really know is this: In a production environment, with an IMAP
> server, given the way most companies use email these days, UW-IMAP+mbox is
> orders of magnitude worse then Cyrus.  The difference is enough that an
> mbox-based system basically thrashes to death due to load, while Cyrus
> chugs
> along nicely.
>
>   Contend all you want.  That's what we see in real life.  :-)

And, dagnabbit, I have to agree.  Here's a typical conversation with an
end-user at Cisco (which, until just before I left, was almost 100% mbox):

IT guy: Mr. Enduser, you have to trim your inbox.

Enduser: Huh?  I've only got 100 MB.  Disk space is -cheap-.  Here, I'll
cut a requisition for you to buy 100 GB more.

IT Guy: No, that's not the issue.  We've got -tons- of space.  The problem
is the file locking.  In order for it to do file locking in its own
particular, magical way, it has to copy the WHOLE INBOX to /tmp, do its
thang, then copy it back again.  EVERY SINGLE TIME YOU POP.  [Cisco was
largely POP at this time, too.]  So, say every five minutes, we have to
copy your inbox -- twice -- and then multiply that by the 200 users on
that server, and you see why we need you to delete some messages.

Enduser: Well... that just sounds dumb.

IT Guy: That's life.

Sure, a zillion directory entries -- with its corresponding huge number of
inodes -- is an imperfect solution.  But, it does two things that, IMHO,
override most all other arguments:

1) It almost entirely avoids even the possibility of (application-caused)
data loss.  Just make sure your filesystem can cope with lots of small
files.
2) Almost as important: it won't bring your mailserver to its knees.

Note, however, that I agree with Ben that merely copying it into the same
directory (and, presumably, the same device) would certainly cut down on a
lot of trouble.  Maybe there's a reason they decided against that: beats
me.  But -- for a record third time -- I'll go along with Ben, and say
that it's the results that matter.

$.02,

-Ken