Thread: slow insert query.

Hello,

i'm using postgresql 7.4.7 on a debian system with a 2.00GHz CPU hyperthreaded, 1GB RAM and 160GB
SCSI disk.

We optimize some parameters to increase performance:
shared_buffers=18750
effective_cache_size=62500
wal_buffers=64
default_statistics_target=100
vacuum_mem=100000
deadlock_timeout=2000

With these parameters we saw the CPU load and the disk usage decrease.

But we still have insert that take more that 1s on little table (24000 tupples with 10 rows).

We've got 633 connexions/mn on our all databases (68 databases with less than 10 rows each) which
each make 1 SQL request (simple select, simple insert, simple update, delete and a few jointure).
Is it normal that an insert take 1s for a such table?

Is it a configuration problem?

Is it a problem of access to the disk?

I've made pgbench but i don't find pgbench reference test.

--

--
Julien WICQUART                NEWTECH MULTIMEDIA
Responsable Exploitation        3 ch du Pigeonnier de la Cépière
Tel: +33 5 61 43 14 80            31081 TOULOUSE
Gsm: +33 6 88 17 16 30            http://www.newtech.fr/
Fax: +33 5 61 43 20 11
j.wicquart@newtech.fr

julien WICQUART wrote:
>
> But we still have insert that take more that 1s on little table (24000 tupples with 10 rows).
>
> We've got 633 connexions/mn on our all databases (68 databases with less than 10 rows each) which

How often are you vacuuming?

What does "VACUUM FULL VERBOSE little_table" show for
removable/non-removable row versions?

--
   Richard Huxton
   Archonet Ltd