Re: Performance suggestions?

-----Original Message-----
From: pgsql-general-owner@postgresql.org
[mailto:pgsql-general-owner@postgresql.org] On Behalf Of Harvey, Allan
AC
Sent: Tuesday, December 14, 2004 3:23 PM
To: pgsql-general@postgresql.org
Subject: [GENERAL] Performance suggestions?

Hi all,

I'm after suggestions on how to keep the initial performance
of a system I've put together for longer than 2 minutes.

I have a small table about 20 rows, a constant, that is receiving
about 160 updates per second.
The table is used to share gathered data to other process
asynchronously.
After 5 min it is 12 updates per second.
Performance returns after a vacuum analyse.

I'm using 7.4.5.
This is the table structure
               Table "public.lastscan"
  Column   |            Type             | Modifiers
-----------+-----------------------------+-----------
 pointnum  | integer                     | not null
 parameter | character varying(8)        | not null
 value     | double precision            | not null
 dt        | timestamp without time zone | not null
Indexes:
    "lsindex" btree (pointnum, parameter)

Or should I just stick to saving the data, inserts seem to go on and on,
and
use a different IPC method.
>>
Is 'pointnum' alone enough to uniquely describe the data (or can it be
chosen in such a way as to uniquely describe the data)?  If so, drop
'parameter' from the index.

What happens if you drop the index altogether?

What happens if you change the index type to hashed?

Storing the data in shared memory instead of a table is an option, of
course.
<<