Date: Sun, 11 Dec 2005 15:13:29 -0500 From: Kris Kennaway <kris@obsecurity.org> To: Danial Thom <danial_thom@yahoo.com> Cc: freebsd-questions@freebsd.org, Kris Kennaway <kris@obsecurity.org> Subject: Re: Freebsd Theme Song Message-ID: <20051211201328.GA5652@xor.obsecurity.org> In-Reply-To: <20051211143349.68091.qmail@web33304.mail.mud.yahoo.com> References: <20051210212424.GA80660@xor.obsecurity.org> <20051211143349.68091.qmail@web33304.mail.mud.yahoo.com>
next in thread | previous in thread | raw e-mail | index | archive | help
--tThc/1wpZn/ma/RB Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Sun, Dec 11, 2005 at 06:33:49AM -0800, Danial Thom wrote: > > > But thats not even the point. The point is that the purpose of > > > tearing apart 4.x was to go MP, and MP=20 > > > performance is dismill across the board. > > This statement is simply false. It's actually > > quite funny to read. > Whats "false" about it, Kris? I've quoted it again for you above. When you take your packet bridging blinders off, there are many performance improvements to be measured from SMP kernels (and UP, for that matter) on FreeBSD 6.0 compared to 4.11. Filesystem performance, for one. FreeBSD 6 is 30% faster than 4.11 at filesystem write operations (extracting a large tarball full of small files and many subdirectories) with an amr disk array on the same UP system. On this hardware FreeBSD 4.11 is unable to make effective use of a second CPU on the same test (it's often slightly slower); FreeBSD 6.0 receives a 10-15% boost on this workload from a second CPU (this seems to be limited by hardware access constraints - the amr hardware API does not encourage concurrency). Performance on a benchmark that does a lot of parallel filesystem reads and forks tens of thousands of processes (ports collection INDEX builds) is 25% faster on 6.0 than 5.4, and is about 3 times faster under SMP than UP on a 4-CPU machine. On a 4-CPU amd64 machine running 6.0, concurrent write performance to a md is 2.7 times faster under SMP than UP. On a 14-CPU sparc64 machine it is 6.1 times faster (and it would be higher except the very low memory bandwidth and 400MHz CPU speed cause some of the kernel threads to saturate easily). But of course, Denial tells us that none of this means anything about FreeBSD performance and scalability, because it doesn't help him with the only thing he cares about, which is to sell systems that bridge high-speed networks. Kris --tThc/1wpZn/ma/RB Content-Type: application/pgp-signature Content-Disposition: inline -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.2 (FreeBSD) iD8DBQFDnIhoWry0BWjoQKURAhjtAJ9at9CvAJvObUJ1kXqUNKh3/4x7yACg1LE1 AqEzwfYEISlgtFuQWGUQUPA= =87pa -----END PGP SIGNATURE----- --tThc/1wpZn/ma/RB--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20051211201328.GA5652>