[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20111120230923.GB13757@openwall.com>
Date: Mon, 21 Nov 2011 03:09:23 +0400
From: Solar Designer <solar@...nwall.com>
To: john-users@...ts.openwall.com
Subject: Re: OpenMP not using all threads
On Sun, Nov 20, 2011 at 03:21:16PM -0700, Stephen John Smoogen wrote:
> Most (if not all) Linode systems are virtualized systems
Yes.
> which I believe normally don't do well with hyperthreading.
I think you're wrong about this.
What I think Linode actually does here is limit the customer's use of
CPU time through giving them only 4 out of probably 16 logical CPUs (in
a dual quad-core Xeon machine). Due to hyperthreading, these 4 actually
let the customer use up to approximately 40% of the 8 physical cores'
performance. If they gave the customer 8 logical CPUs, this would allow
this single customer to almost fully (85% or so) use _both_ Xeon CPUs
installed in the machine (as long as there's otherwise-idle CPU time),
which Linode might not want to offer (or not at the same price).
> For high compute
> virtualization you usually need systems with cpu-pinning and such to
> get the performance that you get from bare metal.
CPU pinning is not that important for these machines and software.
Alexander
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
