Shridhar,
I must disagree with "that's is an OS's job".
OSs try to be generic. With databases, we know more about the data
structure.
If a large table could be split (partitioned) based on specific key, we
could expect huge improvements for agregates queries for example when
that key is involved.
Also, DBA must be able to place table on different file systems. They
know more about the application than the OS does.
Even Postgresql has to be told to perform vaccum and analyze.
If the OS had enough intelligence we could trust it to do a good job,
but until then ...
JLL
"Shridhar Daithankar" wrote:
>
> On Tuesday 18 Feb 2003 2:12 am, you wrote:
> > 3. Please suggest us some tips for setting up a big database to acheive
> > maximum performance ?
>
> well, typically postgresql tries to do things that nobody else in the supply
> chain will do. If an OS would take care of mirroring of drives, then
> postgresql won't do it for example.
>
> Large tables/database are limited by IO performance. So you need to have
> maximum IO throughput available for postgresql to perform best.
>
> I suggest you look at RAID configurations. SCSI if better. If not, then even
> software RAID on some OS's like linux should be a good enough option. Of
> course, put two disk on two channels but that's about it.
>
> Basically, postgresql won't do anything out of box to support disk
> partitioning because that's is an OS's job. symlinking is a poor solution
> because say if you recreate an index, postgresql will drop the symlink and
> recreate a file instead.
>
> HTH
>
> Shridhar
>
> ---------------------------(end of broadcast)---------------------------
> TIP 1: subscribe and unsubscribe commands go to majordomo@postgresql.org