Re: How to improve db performance with $7K?

Alex et al.,

I wonder if thats something to think about adding to Postgresql? A setting for multiblock read count like Oracle (Although

|| I would think so, yea. GMTA: I was just having this micro-chat with Mr. Jim Nasby.

having said that I believe that Oracle natively caches pages much more aggressively that postgresql, which allows the OS to do the file caching).

|| Yea...and it can rely on what is likely a lot more robust and nuanced caching algorithm, but...i don't
know enough (read: anything) about PG's to back that comment up.

Alex Turner
netEconomist

P.S. Oracle changed this with 9i, you can change the Database block size on a tablespace by tablespace bassis making it smaller for OLTP tablespaces and larger for Warehousing tablespaces (at least I think it's on a tablespace, might be on a whole DB).

||Yes, it's tspace level.

On 4/19/05, Jim C. Nasby <decibel(at)decibel(dot)org> wrote:
> On Mon, Apr 18, 2005 at 06:41:37PM -0000, Mohan, Ross wrote:
> > Don't you think "optimal stripe width" would be
> > a good question to research the binaries for? I'd
> > think that drives the answer, largely. (uh oh, pun alert)
> >
> > EG, oracle issues IO requests (this may have changed _just_
> > recently) in 64KB chunks, regardless of what you ask for. So when I
> > did my striping (many moons ago, when the Earth was young...) I did
> > it in 128KB widths, and set the oracle "multiblock read count"
> > according. For oracle, any stripe size under 64KB=stupid, anything
> > much over 128K/258K=wasteful.
> >
> > I am eager to find out how PG handles all this.
>
> AFAIK PostgreSQL requests data one database page at a time (normally
> 8k). Of course the OS might do something different.
> --
> Jim C. Nasby, Database Consultant decibel(at)decibel(dot)org
> Give your computer some brain candy! www.distributed.net Team #1828
>
> Windows: "Where do you want to go today?"
> Linux: "Where do you want to go tomorrow?"
> FreeBSD: "Are you guys coming, or what?"
>
> ---------------------------(end of
> broadcast)---------------------------
> TIP 4: Don't 'kill -9' the postmaster
>