[Dovecot] Improving interaction/performance with Mail.app

Sean Kamath kamath at geekoids.com
Wed Aug 2 02:56:52 EEST 2006


Hi.

Before anyone bothers to mention that I'm an idiot, let me say it first.

OK, I now understand the difference between /dev/random and /dev/urandom.

Now I need to figure out what's really going on, since urandom claims it
will *NEVER* block. . .

Sean

On 8/1/2006, "Sean Kamath" <kamath at geekoids.com> wrote:
>
>On July 1, Alan Schmitt wrote:
>>>>> What OS are you using in server side? And what filesystem?
>>>>
>>>> Mac OS X.4.7, with HFS+ as files system.
>>>
>>> I think I've heard before that Dovecot doesn't behave too great with
>>> OSX..
>>
>>This might be the cause. I've had the plan to move my IMAP server to
>>a linux box, this might be the good time.
>>
>>> Hmm. I guess it's time to put this in wiki:
>>>
>>> http://wiki.dovecot.org/Debugging/ProcessTracing
>>
>>Good to know. I just need to find a way to reproduce the problem.
>>
>>Thanks a lot for your help,
>>
>>Alan
>
>I just moved to Dovecot.  I'm looking forward to figuring out how to
>allow users to have access to (old) email in mbox format, an have all
>new mailboxes go to maildirs. . . But I digress.
>
>When we switched, no one complained of performance issues -- except
>myself.
>
>mail.app would seem to stall on caching a message, then report that it
>was reconnecting, and move a bit further, then stall. . .
>
>What apparently was happening was the dovecot was polling for
>/dev/urandom, over and over again, waiting for some random bits, and
>Mail.app would timeout and reconnect.  Turning off SSL made it MUCH
>faster.
>
>Of course, that's not a solution.
>
>So, I hunted the mailing lists, and ran across some notes about dovecot
>using /dev/urandom a little too enthusiastically.  I haven't had a
>chance to grub through the code (hard to do with a 2-year old that
>won't sleep until midnight on many a nights!), and this post is
>somewhat premature, but I'm curious if turning off SSL removes the
>problem.
>
>Of course, switching to a Linux server will (sort of) negate the problem,
>in that apparently Linux is a little faster at finding random bits (I
>guess, I've never looked into it).
>
>My server is Solaris 9.  My client is Mail.app from 10.3.9.
>
>As I grub through the code, I'll try and identify cryptographically
>significant randomness from non-cryptographically significant randomness.
>
>Sean
>


More information about the dovecot mailing list