<<< Date Index >>>     <<< Thread Index >>>

Re: [Mutt] #1216: mutt unusable on large Maildir with frequent



#1216: mutt unusable on large Maildir with frequent message delivery

Old description:

> {{{
> Package: mutt
> Version: 1.3.28-1
> Severity: important
>
> [NOTE: this bug report has been submitted to the debian BTS as
> Bug#148435.
> Please Cc all your replies to 148435@xxxxxxxxxxxxxxx .]
>
> From: Brian Ristuccia <brian@xxxxxxxxxxxxx>
> Subject: mutt unusable on large Maildir with frequent message delivery
> Date: Tue, 28 May 2002 23:49:30 -0400
>
> On a large Maildir with roughly 40,000 messages and a new message
> arriving
> every 30 seconds, Mutt is mostly unusable. On an Intel Celeron 366 with
> 384MB of ram running Linux kernel 2.4.17, it takes about a minute to read
> the mailbox. By the time Mutt finishes, another message has arrived and
> so
> Mutt starts reading the mailbox again from the beginning.
>
> If this is not a kernel bug, then Mutt has a serious scalability issue.
>
> --
> Brian Ristuccia
> brian@xxxxxxxxxxxxx
> bristucc@xxxxxxxxxx
>
> Received: (at submit) by bugs.guug.de; 2 Dec 2002 01:48:08 +0000
> From merlin@xxxxxxxxxxx Mon Dec 02 02:48:08 2002
> Received: from magic.merlins.org
>         ([216.200.201.205] helo=mail1.merlins.org ident=mail)
>         by trithemius.gnupg.org with esmtp (Exim 3.35 #1 (Debian))
>         id 18Ifgu-0003rG-00
>         for <submit@xxxxxxxxxxxx>; Mon, 02 Dec 2002 02:48:08 +0100
> Received: from saroumane.merlins.org ([198.144.206.12]:32900
> helo=gandalf.merlins.org)
>         by mail1.merlins.org with asmtp
>         (Cipher TLSv1:DES-CBC3-SHA:168) (Exim 4.10-mm1 #1 (Debian))
>         id 18Ific-00063Y-00 by authid <gandalf.merlins.org> with
> auth_plain
>         for <submit@xxxxxxxxxxxx>; Sun, 01 Dec 2002 17:49:55 -0800
> Received: from merlin by gandalf.merlins.org with local (Exim 4.10-mm1 #1
> (Debian))
>         id 18Ifib-0005Pn-00 by authid <merlin>
>         for <submit@xxxxxxxxxxxx>; Sun, 01 Dec 2002 17:49:53 -0800
> Date: Sun, 1 Dec 2002 17:49:53 -0800
> From: Marc MERLIN <marc_soft@xxxxxxxxxxx>
> To: submit@xxxxxxxxxxxx
> Subject: Inefficiency of maildir code -> 1.5.0-me.hcache.8 tryout
> Message-ID: <20021202014953.GB18895@xxxxxxxxxxx>
> Mime-Version: 1.0
> Content-Type: text/plain; charset=us-ascii
> Content-Disposition: inline
> User-Agent: Mutt/1.4i
> X-Sysadmin: BOFH
> X-URL: http://marc.merlins.org/
> X-Operating-System: Proudly running Linux 2.4.19-fswan-1.96-evms-1.1.0
> -swsusp-grsec-1.9.7-marc4-up/Debian woody
> X-Mailer: Some Outlooks can't quote properly without this header
>
> Package: mutt
> Version: 1.5
> Severity: normal
>
> mutt completely trashes my machine for several minutes when it rescans a
> Maildir folder with 20-25k messages
>
> 1.5.0-me.hcache.8 helps for the initial scan when mutt is launched, but
> a rescan after mutt has been launched is still just as slow.
>
> Here are details I posted to the list
>
> On Fri, Nov 08, 2002 at 04:59:48PM +0100, Lorens Kockum wrote:
> > Wouldn't this also be affected by the maildir header caching
> > patch?
>
> Ok, let me report back.
> I have  a 24,000 message  maildir folder,  which isn't small,  but isn't
> *that* huge either (especially for maildir)
> I tried mutt 1.4 vs mutt 1.5.0 vs mutt 1.5.0-me.hcache.8.
> The tests were:
> - mutt -f folder, and wait for the counter to reach 24,000
> - wait some more until I can see the index (instant on mutt 1.4)
> - select a message, add with another mutt a message to the opened
>   maildir, and type 'i' in the test mutt to get back to the index.
>
> Note that some of  those tests in real life on  my laptop (800Mhz, 256MB
> Ram, 40G (5400 RPM)) take 4, 5mn or more when everything isn't cached by
> the OS already. I ran these tests a  few times until the number had been
> lowered enough and my hard drive light pretty much stayed off.
>
> Times are from t=0, i.e. are not cumulative
>
>                         mutt 1.4        mutt 1.5.0      mutt
> 1.5.0-me.hcache.8
>                                                         test #1 test #2
> scanning 24000 messages 30 sec          10 sec          10 sec  8 sec
> giving the index        30 sec          30 sec          120 sec 10 sec
> rescanning maildir      180 sec         90 sec          90 sec  90 sec
>
> So, as you  can see, a maildir rescan  after I add a message  to an open
> folder is still unacceptably slow, although it's indeed twice as fast in
> mutt 1.5.0
>
> mutt 1.5.0-me.hcache.8  took a  long time  to open  my folder  the first
> time, but that's because it was building the index. The following times,
> it opened it in a record 8 to 10 seconds
>
> So Michael's  new code makes  a great  difference for opening  a maildir
> folder once the  index has been built, unfortunately, it  still takes 90
> secs, or  3 times what it  would take to  kill mutt and restart  it, for
> mutt to resync its index with a maildir folder that was modified.
>
> It looks like we're moving in the right direction though.
>
> Let me know if you'd like me to report back on other patches.
>
> Thanks,
> Marc
> --
> "A mouse is a device used to point at the xterm you want to type in" -
> A.S.R.
> Microsoft is to operating systems & security ....
>                                       .... what McDonalds is to gourmet
> cooking
> Home page: http://marc.merlins.org/   |   Finger marc_f@xxxxxxxxxxx for
> PGP key
>

> >How-To-Repeat:
>
> >Fix:
> }}}

New description:

 {{{
 Package: mutt
 Version: 1.3.28-1
 Severity: important

 [NOTE: this bug report has been submitted to the debian BTS as Bug#148435.
 Please Cc all your replies to 148435@xxxxxxxxxxxxxxx .]

 From: Brian Ristuccia <brian@xxxxxxxxxxxxx>
 Subject: mutt unusable on large Maildir with frequent message delivery
 Date: Tue, 28 May 2002 23:49:30 -0400

 On a large Maildir with roughly 40,000 messages and a new message arriving
 every 30 seconds, Mutt is mostly unusable. On an Intel Celeron 366 with
 384MB of ram running Linux kernel 2.4.17, it takes about a minute to read
 the mailbox. By the time Mutt finishes, another message has arrived and so
 Mutt starts reading the mailbox again from the beginning.

 If this is not a kernel bug, then Mutt has a serious scalability issue.

 --
 Brian Ristuccia
 brian@xxxxxxxxxxxxx
 bristucc@xxxxxxxxxx

 Received: (at submit) by bugs.guug.de; 2 Dec 2002 01:48:08 +0000
 From merlin@xxxxxxxxxxx Mon Dec 02 02:48:08 2002
 Received: from magic.merlins.org
         ([216.200.201.205] helo=mail1.merlins.org ident=mail)
         by trithemius.gnupg.org with esmtp (Exim 3.35 #1 (Debian))
         id 18Ifgu-0003rG-00
         for <submit@xxxxxxxxxxxx>; Mon, 02 Dec 2002 02:48:08 +0100
 Received: from saroumane.merlins.org ([198.144.206.12]:32900
 helo=gandalf.merlins.org)
         by mail1.merlins.org with asmtp
         (Cipher TLSv1:DES-CBC3-SHA:168) (Exim 4.10-mm1 #1 (Debian))
         id 18Ific-00063Y-00 by authid <gandalf.merlins.org> with
 auth_plain
         for <submit@xxxxxxxxxxxx>; Sun, 01 Dec 2002 17:49:55 -0800
 Received: from merlin by gandalf.merlins.org with local (Exim 4.10-mm1 #1
 (Debian))
         id 18Ifib-0005Pn-00 by authid <merlin>
         for <submit@xxxxxxxxxxxx>; Sun, 01 Dec 2002 17:49:53 -0800
 Date: Sun, 1 Dec 2002 17:49:53 -0800
 From: Marc MERLIN <marc_soft@xxxxxxxxxxx>
 To: submit@xxxxxxxxxxxx
 Subject: Inefficiency of maildir code -> 1.5.0-me.hcache.8 tryout
 Message-ID: <20021202014953.GB18895@xxxxxxxxxxx>
 Mime-Version: 1.0
 Content-Type: text/plain; charset=us-ascii
 Content-Disposition: inline
 User-Agent: Mutt/1.4i
 X-Sysadmin: BOFH
 X-URL: http://marc.merlins.org/
 X-Operating-System: Proudly running Linux 2.4.19-fswan-1.96-evms-1.1.0
 -swsusp-grsec-1.9.7-marc4-up/Debian woody
 X-Mailer: Some Outlooks can't quote properly without this header

 Package: mutt
 Version: 1.5
 Severity: normal

 mutt completely trashes my machine for several minutes when it rescans a
 Maildir folder with 20-25k messages

 1.5.0-me.hcache.8 helps for the initial scan when mutt is launched, but
 a rescan after mutt has been launched is still just as slow.

 Here are details I posted to the list

 On Fri, Nov 08, 2002 at 04:59:48PM +0100, Lorens Kockum wrote:
 > Wouldn't this also be affected by the maildir header caching
 > patch?

 Ok, let me report back.
 I have  a 24,000 message  maildir folder,  which isn't small,  but isn't
 *that* huge either (especially for maildir)
 I tried mutt 1.4 vs mutt 1.5.0 vs mutt 1.5.0-me.hcache.8.
 The tests were:
 - mutt -f folder, and wait for the counter to reach 24,000
 - wait some more until I can see the index (instant on mutt 1.4)
 - select a message, add with another mutt a message to the opened
   maildir, and type 'i' in the test mutt to get back to the index.

 Note that some of  those tests in real life on  my laptop (800Mhz, 256MB
 Ram, 40G (5400 RPM)) take 4, 5mn or more when everything isn't cached by
 the OS already. I ran these tests a  few times until the number had been
 lowered enough and my hard drive light pretty much stayed off.

 Times are from t=0, i.e. are not cumulative

                         mutt 1.4        mutt 1.5.0      mutt
 1.5.0-me.hcache.8
                                                         test #1 test #2
 scanning 24000 messages 30 sec          10 sec          10 sec  8 sec
 giving the index        30 sec          30 sec          120 sec 10 sec
 rescanning maildir      180 sec         90 sec          90 sec  90 sec

 So, as you  can see, a maildir rescan  after I add a message  to an open
 folder is still unacceptably slow, although it's indeed twice as fast in
 mutt 1.5.0

 mutt 1.5.0-me.hcache.8  took a  long time  to open  my folder  the first
 time, but that's because it was building the index. The following times,
 it opened it in a record 8 to 10 seconds

 So Michael's  new code makes  a great  difference for opening  a maildir
 folder once the  index has been built, unfortunately, it  still takes 90
 secs, or  3 times what it  would take to  kill mutt and restart  it, for
 mutt to resync its index with a maildir folder that was modified.

 It looks like we're moving in the right direction though.

 Let me know if you'd like me to report back on other patches.

 Thanks,
 Marc
 --
 "A mouse is a device used to point at the xterm you want to type in" -
 A.S.R.
 Microsoft is to operating systems & security ....
                                       .... what McDonalds is to gourmet
 cooking
 Home page: http://marc.merlins.org/   |   Finger marc_f@xxxxxxxxxxx for
 PGP key


 >How-To-Repeat:

 >Fix:
 }}}

Comment (by brendan):

 See also #1931

-- 
Ticket URL: <http://dev.mutt.org/trac/ticket/1216#comment:4>