Re: Hardware suggestions

christian(dot)braun(at)tudor(dot)lu wrote:
> Hi list members,
>
> I have a question regarding hardware issues for a SDI (Spatial data
> infrastructure). It will consist of PostgreSQL with PostGIS and a UMN
> Mapserver/pmapper set up.
> At our institute we are currently establishing a small GIS working
> group. The data storage for vector data should be the central PostGIS
> system. Raster data will be held in file system.
> Mostly the users are accessing the data base in read only mode. From
> the client side there is not much write access this only will be done
> by the admin of the system to load new datasets. A prototype is
> currently running on an old desktop pc with ubuntu dapper - not very
> powerfull, of course!
> We have about 10000 € to spend for a new server including the storage.
> Do you have any recommendations for us?
> I have read a lot of introductions to tune up PostgreSQL systems.
> Since I don't have the possibility to tune up the soft parameters like
> cache, mem sizes etc., I wondered about the hardware. Most things were
> about the I/O of harddisks, RAM and file system. Is the filesystem
> that relevant? Because wo want to stay at Ubuntu because of the
> software support, espacially for the GIS-Systems. I think we need at
> least about 300-500Gb for storage and the server you get for this
> price are about two dualcore 2.0 - 2.8 GHz Opterons.
> Do you have any suggestions for the hardware of a spatial data base in
> that pricing category?

Pay as much attention to your disk subsystem as to your CPU / memory
setup. Look at RAID-5 or RAID-10 depending on which is faster for your
setup. While RAID-10 is faster for a system seeing plenty of updates,
and a bit more resiliant to drive failure, RAID-5 can give you a lot of
storage and very good read performance, so it works well for reporting /
warehousing setups.

It might well be that a large RAID-10 with software RAID is a good
choice for what you're doing, since it gets good read performance and is
pretty cheap to implement. If you're going to be doing updates a lot,
then look at a battery backed caching controller.

Memory is a big deal. As much as you can reasonably afford to throw at
the system.

The file system can make a small to moderate impact on performance. Some
loads are favored by JFS, others by XFS, and still others by ext2 for
the data portion (only the pg_xlog needs to be on ext3 meta journaling only)