Re: [Exim] Washington mbx again

On Mon, 2 Oct 2000, Paul Robinson wrote:

> Maildir will also generally be faster on delivery, pickup, use less memory on

Maildir is quite slow if the mailbox contains many different messges.
The directory has be opened, and each file (message) must be opened to get
the header information. Maildir is particularly bad for IMAP, where mail
is left in the mailbox.

MBX which is basically a indexed mbox format is much better. Opening a
MBX mailbox with a 1000 messages is much faster than opening MailDir
mailbox with a 1000 messages.

> large spools, etc. so in general is a Good Thing, but you have to remember
> that your users need to be able to pick the mail up, so if they are POP3'ing
> in, they you are limited to the POP3 daemons you can use (personally, I would
> pay large wedges of cash for qpopper to be able to do maildir natively), and
> the same can be said for any mail pickup system - delivering mail to the mail
> spool on the server is only half the battle. ;-)
>
> > We also backup everything using Legato, and if each mail occupies one
> > file, then the index will be very very big. On the other hand mbx
> > format will just have one file/user and the index will not be that big.
> > I have noticed that using standard Unix mailbox format, swap space used
> > by individual user is directly related to the mailbox (inbox/folder)
> > size. Is the same with mbx format? How does maildir uses the swap
> > space?
>
> Hmmm. Perhaps I am mis-reading something here, but I can't see how swap space
> usage is determined from mailbox usage. I take it we're talking about proper

The UW pop server, when used with mbox format mailboxes, loads the
entire mailbox into memory. This uses lots of VM. The entire mailbox
file needs to be read to determine the number of messages present, so the
UW pop server keeps the mail resident. It is memory vs. performance
tradeoff. I understand the newer versions do something different. It was
original done for locking. The mbox format is crappy, so POP and IMAP
servers must do all kinds of things to avoid stepping on the MTA.

> Unix virtual memory, and not just some files being moved around? Swap space
> usage is determined by the VM policy of the underlying operating system - for
> example, Linux tries to avoid using swap at all costs, *BSD will swap out
> processes that have been idle for "X" amount of time, etc.
>
> Or am I missing something obvious here? (I've got flu, OK?) :-)


Tom