>>>>> "Thomas" == Thomas Roessler <[EMAIL PROTECTED]> writes:
>>>>> "Lars" == Lars Hecking <[EMAIL PROTECTED]> writes:
Thomas> You are transferring almost 8 MBit/s with the new mutt versions.
Thomas> This looks like the bottleneck is really NFS and your Ethernet,
Thomas> not mutt. With the old mutt, you get more than 34 MBit/s, which
Thomas> looks like your system's local NFS cache.
Thomas> -HOMESPOOL -USE_SETGID +USE_DOTLOCK +USE_FCNTL -USE_FLOCK
Thomas> This output confirms Brendan's suspicion. More precisely, you
Thomas> have one of the few system configurations which get into locking
Thomas> hell with old mutt versions.
Thomas> What happens is this: Your system has a mail spool which doesn't
Thomas> require additional privileges to set up (-USE_SETGID). This means
Thomas> that dotlocking (+USE_DOTLOCK) is done within mutt, and not in an
Thomas> external program. However, this actually breaks fcntl locking by
Thomas> opening and closing the spool file _after_ doing the fcntl lock,
Thomas> thereby actually releasing it.
It sounds like you are saying that I should change the combination of
+USE_DOTLOCK +USE_FCNTL -USE_FLOCK
to something else but it isn't clear to me what I should change it to.
Are there any changes I should make to the HP-UX configuration. That has
+USE_SETGID +USE_DOTLOCK +USE_FCNTL -USE_FLOCK
Thomas> Note that maildir folders may actually lead to a performance
Thomas> improvement for you: This folder format doesn't require any
Thomas> locking, that is, mutt will fully benefit from the NFS client's
Thomas> cache.
I'm not sure that I want to do that as I sometimes use Netscape to read mail
folders.
An additional point of information.
My mail is really downloaded with fetchmail from a POP server and then
processed with procmail. On occasion, mail may be delivered to my
workstation's sendmail. This should be no different than fetchmail's action
(since it delivers to port 25 and sendmail takes it from there).
Lars> I had an exchange about mutt 1.1.x slowness with tlr two weeks ago.
Lars> I noticed the same slowdown after version 1.1.2. The slowdown
Lars> occured even for small (a few dozen kB), local mailboxes. The
Lars> explanation: I fixed charmap support, i.e. finally installed the
Lars> charmaps-0.0 stuff into /usr/local/share/mutt/charmaps (/usr/local
Lars> is mounted across the network).
Lars> The slowdown rate is proportional to the number of different
Lars> Content-Type/charset= headers in a mailbox file, apparently because
Lars> the charmaps are loaded along with the messages.
Lars> If you leave the same instance of mutt open and (c)hange mailboxes,
Lars> the charsets should be cached.
I usually do leave a single instance open.
Lars> <rant> Convenient as they are, charsets are another feature that
Lars> make it easier for ppl to shoot themselves (and others) in the
Lars> foot. Now that my mutt is charset sensitive, I often find messages
Lars> with big5, iso-2022-jp, or koi8-r, although none of the special
Lars> characters are actually in the message body (checked with vi in
Lars> "C"). Free performance killer with every mail message. </rant>
Is there a way to diable this "feature"?
--
Eric M. Boehm [EMAIL PROTECTED]
