On 2001-08-22 17:01:40 +0200, Alexander Skwar wrote:
>Also the load when opening the Maildir is WAY higher compared to
>the mbox file. Load when opening the mbox is about 2, Maildir is
>about 6. Okay, I've got setiathome running, so one may substract 1
>(? right?).
Ouch. PLEASE make sure that (1) swapping isn't necessary, (2) your
CPU is mostly idle when you do measurements, (3) mutt (or, for that
matter, evolution) is the only process which competes for disk
access.
>Opening the Maildir took close to 5 minutes (the first time), but
>at least the load only went up to 5 and the system also felt more
>responsible.
That's interesting - seems to translate to less disk accesses per
unit time, which could mean that evolution is a bit slower at
opening maildir folders.
(Bad enough, I can't make any reasonable measurements myself - I
don't have Evolution installed, it's not available as a package for
the distribution I use, and, finally, maildir is mostly kernel-bound
here - at a certain point, ext2 eats most of the CPU.)
>The 2nd time it took 15 seconds (!!). I don't know, but judging
>this tremendous speedup, I suppose Evolution is keeping a "cache"
>of somesorts which causes it not to read the whole directory.
>That's just too much of a speedup, I'd say.
Seems so.
Now, such a cache will have to introduce explicit locking with
maildir folders. But the entire point with this folder format is
that you do not have to lock explicitly, because the locking which
is needed happens in the filesystem, when it guarantees that
directories are consistent.
>In mutt, changing from the default date/date sort to subject/date
>takes, uhm, 1 second. But enabling threads (thread/date) takes
>MUCH longer, to be "exact", it took about 2 minutes 30 seconds.
Try a newer version. Someone has contributed a patch which improves
mutt's threading algorithm to O( n log n ); it is in mutt-1.3.21.
--
Thomas Roessler http://log.does-not-exist.org/
