* On 2004.01.20, in <[EMAIL PROTECTED]>,
* "Clifton Royston" <[EMAIL PROTECTED]> wrote:
>
> Usually this kind of overload is due to many users having large
> mailboxes (e.g. 30MB and up) in the old UNIX mbox format. In this
Bart -
I've never run qpopper with the Maildir patch, so I don't have an
automatic sense of how to compare. And I'm not sure how much you should
expect from your particular server system with Maildir, either. I can
say, though, that once you start to hit the limits of your I/O system,
performance degrades very quickly. It's no longer linear, it's like you
take your foot off the accelerator, and the brakes come on in full.
I've been putting off a reply to see what others say, but I might
as well go ahead with mentioning that if it does seem to be a high
user-load problem, or if that seems like a good-enough temporary
solution, you might want to take a look at how we reduced that at my
site:
http://home.uchicago.edu/~dgc/sw/qpopper/index.html
(I still have some contributed patches to apply to this -- for which I
apologize to their submitters -- but they don't affect basic operation.)
When our server first hit its I/O limits, we saw pretty much three
options: divide onto multiple servers, speed up the storage subsystem,
or slow down the users. The first two weren't good solutions for us at
the time, so we opted for forcing the users to stop hitting the server
so hard. Since then we've improved our I/O capacity and we will be
splitting onto multiple servers, but the rate-limiting we levied back
then helped us through the rough period.
--
-D. [EMAIL PROTECTED] ** Enterprise Network Servers and Such
** University of Chicago
We are the robots. ** North America's southernmost seasonal glacier
