[postgis-users] Rv: Postgis for Postgres 64 bits

[Andy Colson]

On 02/04/2012 01:14 PM, Darrel Maddy wrote:
> Dear Regina,
>
> I would be happy to contribute $100 into any fund which secures a 64-bit windows build of PostGIS2 -especially as I am about to install it on a machine with 64Gb RAM :)
>
> Count now at 1 ?
>
> Thanks
>
> Darrel
> http://www.ncl.ac.uk/gps/staff/profile/darrel.maddy#tab_profile
>
>


I really wonder if its faster.  I'd bet not.  If you consider the architecture, I'd bet, for all but the very largest of datasets, 32 bit PG running on top 64 bit os is faster.

Consider: you want the os disk cache to be as large as possible.  You want a bunch of shared mem and I'd bet around 2 gig would be a sweet spot.  More shared mem wouldn't necessarily be a benefit. (There are upper bounds recommended for shared_buffers, especially on windows).

PG spawns a separate client for each connection, so each one of them could access up to 2 gig.  (so things like work_mem could be large).

And last, 32bit programs are faster because they move less data around (32bit pointers instead of 64bit).

Taken all together PostMaster + (clients * postgres.exe) + os cache could easly be more than 2 gig using 32 bit PG on 64 bit windows.

The only benefit of using 64bit PG would be a larger shared mem.  The online manual says: "The useful range for shared_buffers on Windows systems is generally from 64MB to 512MB"

http://www.postgresql.org/docs/9.1/static/runtime-config-resource.html

The best use of ram, is as a disk cache, which 64 bit windows should use as much as possible (well, I assume).  After that there is nothing in Postgres (Per Client) that is going to run faster with more than 2 gig of ram.

Has anyone ever done any benchmarks?

-Andy