Re: Performance: files or SQL?

[Justin Mason <jm...@jmason.org>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


Michael Parker writes:
> Cami wrote:
> 
> > Well.. As for setup, we're using amavisd-new which provides more
> > functionality than spamc/spamd (thus producing more sql queries).
> > As for average scantime per msg, no idea since i dont log that.
> > Hardware wise: Dell PowerEdge 2650, Dual 3.06GHz, 6gigs of ram,
> > 5x32K RPM disks in Raid 10/1+0..
> 
> Ahhh....so the more correct statement would be amavisd-new with SQL does
> not scale.  Since I assume you are using a global sitewide database with
> amavis this doesn't surprise me.  There would be a lot of lock
> contention with that many children attempting to hit a single row in the
> database constantly.  A better setup would be individual bayes dbs and
> InnoDB tables (which support row level locking).  That would certainly
> scale much better.

Is there any diff between Amavisd and spamd's use of SQL for bayes?
I wouldn't have thought so...

- --j.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)
Comment: Exmh CVS

iD8DBQFC0/WYMJF5cimLx9ARAqs0AJ0bOPkDbcX9T6xtrJkAeGJcwRtPOgCfW7m4
0X2fxfMtNfFkHkBOTAB5OY0=
=QNgY
-----END PGP SIGNATURE-----

Re: Performance: files or SQL?

[Michael Parker <pa...@pobox.com>]

Justin Mason wrote:

> Is there any diff between Amavisd and spamd's use of SQL for bayes?
> I wouldn't have thought so...


No, just the fact that it doesn't generally support (someone feel free
to correct me if I'm wrong) individualized configs (ie everything runs
as a single user).  Anyone who attempts a sitewide database with that
many children hitting the db at once will most likely encounter the same
sort of difficulties.  However, 4000 queries/sec is not very high (I can
easily push that via 2 spamd children), so I'm actually guessing there
is a broader issue here.  Maybe amavisd is doing some additional, poorly
optimized, queries that is causing the database slowness.

Michael