Re: Postgres & large tables on average machine - Mailing list pgsql-admin

From Jeremy Buchmann
Subject Re: Postgres & large tables on average machine
Date
Msg-id 3CA513E2.5000501@wellsgaming.com
Whole thread Raw
In response to Re: Postgres & large tables on average machine  ("Fred Moyer" <fred@digicamp.com>)
Responses Re: Postgres & large tables on average machine  ("Nicholay P. Chuprynin" <kolyan@infoport.uz>)
List pgsql-admin
Fred Moyer wrote:
> Your UDMA 33 bus will limit disk reads to 33 Mbytes/sec so there is your
> first bottleneck.  Get a  66 mhz PCI ide adapter (Promise is cheap) and that
> will increase your disk speed dramatically.
> Also you won't be able to do much with 128 Mb of ram, put in as much as you
> can.  That box is likely 66 mhz front side bus so that will be a bottleneck
> once you max out the ram.
 >
[snip]
>
>>Resently I had to create and manage the (relatively) large table.
>>In the mean time it's about 8 million rows, and surely will grow above
>>this size.
>>The problem is that queries takes absolutely not acceptable time.
>>Database located on average Celeron 400 machine with 128 Mb of RAM and
>>UDMA 33 capable IDE drive.
>>I run PostgreSQL 7.1 on Debian Linux with 2.4.18 kernel.
>>My question is what could be done in order to improve the performance?
>>I mean, is that normal behavior for Postgres on such computer or I
>>encounter a misconfiguration?

Also, the Celeron processor is cache-starved.  If you can switch it out
for a P3 at even the same clock speed, it'd be worth it.  A while back,
someone posted benchmarks where he just changed the processor from a
Celeron to a P3 of the same clock speed and the P3 was twice as fast.

Databases are I/O bound.  Anything you can stuff into cache is worth it.
This is why lots of memory, big disk caches, disk controller caches, and
processor caches help so much.

--Jeremy


pgsql-admin by date:

Previous
From: "Fred Moyer"
Date:
Subject: Re: Postgres & large tables on average machine
Next
From: MG
Date:
Subject: Re: PgSQL postmaster.opts