Subject: Re: GPU Accelerated Sorting
From: Jeff Davis <pgsql@j-davis.com>
To: Eliot Gable <egable+pgsql-performance@gmail.com>
Cc: pgsql-performance@postgresql.org
In-Reply-To: <AANLkTimjYxrvMGMLhh3q=82iNiKKjQvPRcrsgX7DHcMp@mail.gmail.com>
References: <AANLkTimjYxrvMGMLhh3q=82iNiKKjQvPRcrsgX7DHcMp@mail.gmail.com>
Content-Type: text/plain; charset="ISO-8859-1"
Date: Mon, 30 Aug 2010 15:01:47 -0700
Message-ID: <1283205707.6656.15.camel@jdavis-ux.asterdata.local>
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit

On Mon, 2010-08-30 at 09:51 -0400, Eliot Gable wrote:
> Not sure if anyone else saw this, but it struck me as an interesting
> idea if it could be added to PostgreSQL. GPU accelerated database
> operations could be very... interesting. Of course, this could be
> difficult to do in a way that usefully increases performance of
> PostgreSQL, but I'll leave that up to you guys to figure out.
> 
> http://code.google.com/p/back40computing/wiki/RadixSorting
> 

Radix sort is not a comparison sort. Comparison sorts work for any data
type for which you define a total order; and any total order is allowed.
Radix sort only works for some data types and some total orders.

However, it would be very nice to use radix sorting where it does work.
That would require some extensions to the type system, but it could be
done.

The GPU issue is orthogonal.

Regards,
	Jeff Davis