Thread: Page Size in Future Releases

Page Size in Future Releases

From
"Kendrick C. Wilson"
Date:
Will a increase in the size of a data page increase performance of a
database with large records?

I have records about 881 byte + 40 byte (header) = 921.

8k page size / 921 bytes per record is ONLY 8 records...........

Comments are welcome.........

k=n^r/ck, SCJP

_________________________________________________________________
MSN 8 with e-mail virus protection service: 2 months FREE*
http://join.msn.com/?page=features/virus


Re: Page Size in Future Releases

From
"Shridhar Daithankar"
Date:
On Friday 21 Mar 2003 2:15 am, Kendrick C. Wilson wrote:
> Will a increase in the size of a data page increase performance of a
> database with large records?
>
> I have records about 881 byte + 40 byte (header) = 921.
>
> 8k page size / 921 bytes per record is ONLY 8 records...........

You can tweak it yourself at compile time in some header file and that should
work but that is a point of diminising results as far as hackers are
concerned.

One reason I know where it would help is getting postgresql to use tons of
shaerd memory. Right now postgresql can not use much beyond 250MB(??) because
number of shared buffer are int or something. So if you know your reconrds
are large, are often manipulated and your OS is not so good at file caching,
then increasing page size might help.

Given how good unices are in general in terms of file and memory handling, I
woudl say you should not do it unless your average record size is greater
than 8K, something like a large genome sequence or so.

YMMV..

 Shridhar

Re: Page Size in Future Releases

From
"Christopher Kings-Lynne"
Date:
> > I have records about 881 byte + 40 byte (header) = 921.
> >
> > 8k page size / 921 bytes per record is ONLY 8 records...........
>
> You can tweak it yourself at compile time in some header file and that
should
> work but that is a point of diminising results as far as hackers are
> concerned.

As far as I'm aware the 8k page size has nothing to do with speed and
everything to do with atomic writes.  You can't be guaranteed that the O/S
and hard drive controller will write anything more than 8K in an atomic
block...

Chris


Re: Page Size in Future Releases

From
Josh Berkus
Date:
Shridar,

> One reason I know where it would help is getting postgresql to use tons of
> shaerd memory. Right now postgresql can not use much beyond 250MB(??)
> because number of shared buffer are int or something. So if you know your
> reconrds are large, are often manipulated and your OS is not so good at
> file caching, then increasing page size might help.

Um, two fallacies:
1) You can allocate as much shared buffer ram as you want.  The maxium I've
tested is 300mb, personally, but I know of no hard limit.

2) However, allocating more shared buffer ram ... in fact anything beyond
about 40mb ... has never been shown by anyone on this list to be helpful for
any size database, and sometimes the contrary.


--
Josh Berkus
Aglio Database Solutions
San Francisco

Re: Page Size in Future Releases

From
Tom Lane
Date:
"Kendrick C. Wilson" <kendrick_wilson@hotmail.com> writes:
> Will a increase in the size of a data page increase performance of a
> database with large records?

Probably not; in fact the increased WAL overhead could make it a net
loss.  But feel free to try changing BLCKSZ to see how it works for you.

            regards, tom lane

Re: Page Size in Future Releases

From
Mario Weilguni
Date:
Am Samstag, 22. März 2003 01:15 schrieb Tom Lane:
> "Kendrick C. Wilson" <kendrick_wilson@hotmail.com> writes:
> > Will a increase in the size of a data page increase performance of a
> > database with large records?
>
> Probably not; in fact the increased WAL overhead could make it a net
> loss.  But feel free to try changing BLCKSZ to see how it works for you.

I've several database with 32KB and 8KB, and though the results are not really comparable due to slight different
hardware,I've the feeling that 8KB buffers work best in most cases. The only difference I noticed are large objects
whichseem to work slightly better with larger sizes. 

Regards,
    Mario Weilguni