Thread: Hyper-Threading

Hyper-Threading

From
lista@profit-ti.com.br
Date:
Hi People! 
As a project manager I decided to migrate the full project from Oracle 
to PostgreSQL.

We bought a new Intel box to host the database, it is a dual Xeon 2.2 
and I intend to use it with a FreeBSD 4.8  wich support Intel Hyper-
Threading.
I Woul like to know if the PostgreSQL 7.3.2 will take advantage from 
Hyper-Threading?
Best Regards,



Re: Hyper-Threading

From
Doug McNaught
Date:
lista@profit-ti.com.br writes:

> Hi People! 
> As a project manager I decided to migrate the full project from Oracle
> to PostgreSQL.
> 
> We bought a new Intel box to host the database, it is a dual Xeon 2.2
> and I intend to use it with a FreeBSD 4.8  wich support Intel Hyper-
> Threading.
>  
> I Woul like to know if the PostgreSQL 7.3.2 will take advantage from
> Hyper-Threading?

The question is really the same as "will PostgreSQL take advantage of
multiple processors?"  The answer is that it will not split single
queries across multiple virtual or physical processors, but if you
have multiple clients connected each client backend (which is an
independent process) can be scheduled on any CPU available to it.

-Doug



Re: Hyper-Threading

From
"scott.marlowe"
Date:
On Thu, 8 May 2003 lista@profit-ti.com.br wrote:

> Hi People! 
> As a project manager I decided to migrate the full project from Oracle 
> to PostgreSQL.
> 
> We bought a new Intel box to host the database, it is a dual Xeon 2.2 
> and I intend to use it with a FreeBSD 4.8  wich support Intel Hyper-
> Threading.
>  
> I Woul like to know if the PostgreSQL 7.3.2 will take advantage from 
> Hyper-Threading?

Postgresql is not a multi-threaded app, so my guess is no, it won't.  but 
if the OS benefits from it then you may see a slight speed increase.

while a neat idea, hyperthreading seems to be promising far more than it 
is delivering.

In testing on Windows XP with SQL server, the performance increase on a 
single CPU machine with hyperthreading enabled was only about 20%.  Of 
course, Microsoft charges you for a whole nother CPU for that 20% 
performance gain. :-)

The box Being dual CPU will make a bigger difference than being 
hyperthreaded.



Re: Hyper-Threading

From
The Hermit Hacker
Date:
I've heard conflicting reports about HyperThreading in general ... by
default, FreeBSD doesn't enable it (it originally did), as ppl were
reporting performance degradation ... I'm using it on our servers, and
haven't noticed any problems ...

As far as PostgreSQL is concerned, how well it takes advantage of it is
alot more dependent, I think, on the OS itself then on the application ...
PostgreSQL is multi-process, but its the OS itself that handles the
schedualing between CPUs, and how those resources are allocated ...

My understanding of HyperThreading is that on a DualCPU box, you end up
with 4 CPUs running at ~67% each, instead of 2 CPUs running at 100% each
... or ~1.5x faster, depending on how well the OS itself handles it ...

On Thu, 8 May 2003 lista@profit-ti.com.br wrote:

> Hi People! 
> As a project manager I decided to migrate the full project from Oracle
> to PostgreSQL.
> 
> We bought a new Intel box to host the database, it is a dual Xeon 2.2
> and I intend to use it with a FreeBSD 4.8  wich support Intel Hyper-
> Threading.
>  
> I Woul like to know if the PostgreSQL 7.3.2 will take advantage from
> Hyper-Threading?
> Best Regards,
>
>
> ---------------------------(end of broadcast)---------------------------
> TIP 1: subscribe and unsubscribe commands go to majordomo@postgresql.org
>
>

Marc G. Fournier                   ICQ#7615664               IRC Nick: Scrappy
Systems Administrator @ hub.org
primary: scrappy@hub.org           secondary: scrappy@{freebsd|postgresql}.org



Re: Hyper-Threading

From
Sean Chittenden
Date:
> I've heard conflicting reports about HyperThreading in general
> ... by default, FreeBSD doesn't enable it (it originally did), as
> ppl were reporting performance degradation ... I'm using it on our
> servers, and haven't noticed any problems ...

Twiddle with the sysctl machdep.cpu_idle_hlt, that seemed to make a
big difference in terms of power consumption, heat, and effectiveness
of HTT.  As the previous poster suggested, HTT promises lots, but
delivers little.  Search the archives on FreeBSD.org for a set of
rather large threads regarding this topic.  -sc

-- 
Sean Chittenden