Re: SSD options, small database, ZFS - Mailing list pgsql-performance

From Arjen van der Meijden
Subject Re: SSD options, small database, ZFS
Date
Msg-id 4EC602F6.5030105@tweakers.net
Whole thread Raw
In response to Re: SSD options, small database, ZFS  (CSS <css@morefoo.com>)
List pgsql-performance

On 18-11-2011 4:44 CSS wrote:
> Resurrecting this long-dormant thread...
>> Btw, the 5500 and 5600 Xeons are normally more efficient with a multiple of 6 ram-modules, so you may want to have a
lookat 24GB (6x4), 36GB (6x4+6x2) or 48GB (12x4 or 6x8) RAM. 
>
> Thanks - I really had a hard time wrapping my head around the rules on populating the banks.  If I understand it
correctly,this is due to the memory controller moving from the south(?)bridge to being integrated in the CPU. 

That's not complete. A while back Intel introduced an integrated memory
controller in the Xeon's (I think it was with the 5500). And doing so,
they brought NUMA to the mainstream Xeons (Opterons had been doing that
from the start).
The memory controllers in 5500/5600 are "triple channel". I.e. they can
distribute their work over three memory channels at the same time. The
next generation E5 Xeon's will have "quad channel", so it'll be going
even faster with module count than.

With these kinds of cpu's its normally best to have increments of "num
channels"*"num cpu" memory modules for optimal performance. I.e. with
one "triple channel" cpu, you'd increment with three at the time, with
two cpu's you'd go with six.

Having said that, it will work with many different amounts of memory
modules, just at a (slight?) disadvantage compared to the optimal setting.

Best regards,

Arjen

pgsql-performance by date:

Previous
From: CSS
Date:
Subject: Re: SSD options, small database, ZFS
Next
From: Greg Matthews
Date:
Subject: probably cause (and fix) for floating-point assist faults on itanium