Subject: Re: Bad count of CPUs -- On Hyper Threading
To: None <current-users@NetBSD.org>
From: gabriel rosenkoetter <firstname.lastname@example.org>
Date: 10/16/2006 12:02:42
Content-Type: text/plain; charset=us-ascii
On Thu, Oct 12, 2006 at 08:56:54PM +0800, Joseph A. Dacuma wrote:
> > Unless your scheduler completely understands HT, and assigns processes
> > accordingly, and unless all the idle loops and busy-wait loops are
> > HT correct, you could easily be better off disabling HT.
> > David
> Thank you very much for all the info on HT. My last question is, does
> NetBSD completely understand HT? Or will one be better off disabling it
> completely since the performance gains is not that glaring compared to the
> risks involved when enabled for common systems.
I think your question there is, "Does the NetBSD scheduler
completely understand HT?" ... in which case the answer is: "The
NetBSD scheduler is completely ignorant of HyperThreading, and you
will probably want to turn it off."
It's worth testing your application to be sure, but chances are good
that you'll either see zero difference (because you're not
CPU-bound, so having twice as many doesn't mean anything) or you'll
periodically get burned by processes sitting on the same physical
processor while another physical processor sits completely idle.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v126.96.36.199 (FreeBSD)
-----END PGP SIGNATURE-----