XEON familiy 5000, 5100 or 5300?

Philippe Lang wrote:

If I'm not wrong, a single postgresql sql query cannot be spread over
two processors, but can it be spread over multiple cores? If that's

No - a *core* is another cpu, basically you will have 2 or 4 cpu's in
the one physical package.

HT creates 2 virtual cpu's sharing the same cpu resources but the cores
are seperate cpu's in themselves.

The Quad-core will only benefit you more if you have more users running
queries at the same time. Each core can run a query at the same time
without slowing the others down (allowing for disk access/FSB limits).

--

Shane Ambler
pgSQL@007Marketing.com

Get Sheeky @ http://Sheeky.Biz

PostgreSQL handles each connection in a dedicated process, so you won't get
better performance for a single connection by adding more CPUs (I mean,
beyond the benefit of having the postmaster and the specific connection
running in separate CPUs). This means that a query will not be resolved by
more than one CPU. What you will get is better performance for multiple
connections.

On Sat January 13 2007 05:43, Philippe Lang wrote:

Hi,

I'm about to buy a few new servers, and I'm trying to determine if I
should buy XEON family 5000, 5100 or 5300 processors.

For about the same price, I can have:

2 Dual-Core Intel Xeon 5060, 3.2 GHz, 4MB
2 Dual-Core Intel Xeon 5130, 2.0 GHz, 4MB
2 Quad-Core Intel Xeon 5310, 1.6 GHz, 4MB

I have a few queries that take about 4 minutes each to complete on a
single Pentium 4, and all of them are CPU-bound, with the whole database
in RAM. With the new system, I expect a performance boost, of course!

If I'm not wrong, a single postgresql sql query cannot be spread over
two processors, but can it be spread over multiple cores? If that's
true, does that mean the best CPU would be the last one, although the
clock is lower that the one of the other processors?

Thanks for the infos,

Cheers,

---------------
Philippe Lang
Attik System

---------------------------(end of broadcast)---------------------------
TIP 6: explain analyze is your friend

--
Juan Jose Comellas
(juanjo@comellas.com.ar)

Shane wrote:

No - a *core* is another cpu, basically you will have 2 or 4 cpu's in
the one physical package.

HT creates 2 virtual cpu's sharing the same cpu resources but the
cores are seperate cpu's in themselves.

The Quad-core will only benefit you more if you have more users
running queries at the same time. Each core can run a query at the
same time without slowing the others down (allowing for disk
access/FSB limits).

Jose wrote:

PostgreSQL handles each connection in a dedicated process, so you
won't get better performance for a single connection by adding more
CPUs (I mean, beyond the benefit of having the postmaster and the
specific connection running in separate CPUs). This means that a
query will not be resolved by more than one CPU. What you will get is
better performance for multiple connections.

Shane, Jose,

Thanks for your answers. In my "very-low-concurrency scenario", I guess
then that multiple cores won't really help, as I suspected.

I think I have better take (for the same price) a ...

Dual-Core Intel Xeon 5060, 3.2 GHz, 4MB

... instead of a ...

Quad-Core Intel Xeon 5310, 1.6 GHz, 4MB

With my CPU-bound query, it will perform better.

But what about Hyperthreading then? Is it able to spread two threads
over two different cores? I guess the answer is no...

Philippe