I'm using a 4 Xeon Hyperthreaded server with Windows 2003 ent. and SQL
Server 2000 ent. (8.0.818)
Whenever a user launch a specific function, let's say a sp, he uses 100% of
one of the 8 CPUs during 30-40 minutes and the workload jump from one
processor to another (every time the sp makes a new query).
I read in here that most of the query, especially the simple ones or the
bulk ones run on only one processor, so I think the behaviour of the sp
described above is normal.
My problem is when a second user comes in and run another application,
running on another database, and run a big process (again a sp doing various
things during 30 minutes or so) I don't see 2 processors running at 100% but
only one, and the process time of the two sp running simultanneously is
almost doubled...
Maybe I have a performancebottleneck somewhere, but I don't see where, I
have some pretty good SCSI Disk, 6GB of Ram almost never used more than 30%.
Except upgrading my processors speed is there a way to have better
performance? Like forcing more parrallelism on some process or queries. I
read something about query plan complexity defining the level of parallelism
of a query, is there a way to tune this level?
Thanks in advance.
Best regard,
Christophe.
There is a setting that adjusts the threshhold for parallism. sp_configure will show you the setting.
Of course indexing well is almost always the best path to follow.
|||"Ken Dutton" <fj60landcruiser@.yahoo.com> a crit dans le message de
news:505E0AD6-A9E0-4BE4-9214-1B3BC7BEE3B8@.microsoft.com...
> There is a setting that adjusts the threshhold for parallism.
sp_configure will show you the setting.
Thanks for your answer,
The threshold is currently at 3. I read in the BOL it's in millisecond, so I
don't think I can make it lower...
But that helped me see I didn't activate the AWE parameter on this server...
I was wondering why I never used more than 2GB!!
> Of course indexing well is almost always the best path to follow.
Yay!
That's what I'm telling my sql devs and db designers every morning, but that
doesn't seem to motivate them... I think a wooden club with long nails will
do the trick, but I'll try to coerce them into indexing their tables and
views one last time...
I'll do some trace and index tuning before, I'm not a monster... yet.
Anyway, thanks again.
Christophe
