Subject: Re: size of cache
From: Simon Riggs <simon@2ndquadrant.com>
To: Jean-Max Reymond <jmreymond@gmail.com>
Cc: pgsql-performance@postgresql.org
In-Reply-To: <4b09a0c05071301207d452ace@mail.gmail.com>
References: <4b09a0c05071301207d452ace@mail.gmail.com>
Content-Type: text/plain
Organization: 2nd Quadrant
Date: Wed, 13 Jul 2005 17:07:47 +0100
Message-Id: <1121270867.3970.266.camel@localhost.localdomain>
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit

On Wed, 2005-07-13 at 10:20 +0200, Jean-Max Reymond wrote:
> with my application, it seems that size of cache has great effect:
> from 512 Kb of L2 cache to 1Mb boost performance with a factor 3 and
> 20% again from 1Mb L2 cache to 2Mb L2 cache.

Memory request time is the main bottleneck in well tuned database
systems, so your results could be reasonable. 

> I don't understand why a 512Kb cache L2 is too small to fit the data's
> does it exist a tool to trace processor activity and confirm that
> processor is waiting for memory ?

You have both data and instruction cache on the CPU. It is likely it is
the instruction cache that is too small to fit all of the code required
for your application's workload mix.

Use Intel VTune or similar to show the results you seek. 

Best Regards, Simon Riggs