On Sat, 2005-07-23 at 08:02 -0400, Kevin Flanagan wrote:
> The up side is performance, while it's not as linear as SMP, it's a
> performance gain just how much depends on your OS and applications. The
> best numbers I have heard are 40% to 60% performance gain.
> All of that said, I would expect this new Intel CPU to appear as 4 to
> the OS and software. I haven't seen a real performance gain from
> hyperthreading, but I haven't done very fine level studying of a system,
> just the "feel of it". Dual core should offer real performance gains for
> some things, but mostly it'll just behave as if it were an SMP system.
So which is it? 40--60% or nothing?
In our testing we've seen only one application (an NFS server) that saw
more than 10% improvement and it was ~15%. And yes, we were using a 2.6
kernel with the HT-aware scheduler.
The vast majority of the applications that we run are memory-bandwidth
limited and, in those cases, HT does basically nothing (all <1% and
sometimes *negative*) to help. If you're one of those (rare?) folks
that isn't memory bandwidth limited, then perhaps you'll see a little
improvement. But for most folks the memory system is the bottleneck and
adding more processors (either real or virtual) without increasing bw is
utterly pointless.
Ed
--
Edward H. Hill III, PhD
office: MIT Dept. of EAPS; Rm 54-1424; 77 Massachusetts Ave.
Cambridge, MA 02139-4307
emails: [EMAIL PROTECTED] [EMAIL PROTECTED]
URLs: http://web.mit.edu/eh3/ http://eh3.com/
phone: 617-253-0098
fax: 617-253-4464
--
TriLUG mailing list : http://www.trilug.org/mailman/listinfo/trilug
TriLUG Organizational FAQ : http://trilug.org/faq/
TriLUG Member Services FAQ : http://members.trilug.org/services_faq/
TriLUG PGP Keyring : http://trilug.org/~chrish/trilug.asc
