From owner-freebsd-hackers Mon Oct 23 12:15:04 1995 Return-Path: owner-hackers Received: (from root@localhost) by freefall.freebsd.org (8.6.12/8.6.6) id MAA19288 for hackers-outgoing; Mon, 23 Oct 1995 12:15:04 -0700 Received: from phaeton.artisoft.com (phaeton.Artisoft.COM [198.17.250.211]) by freefall.freebsd.org (8.6.12/8.6.6) with ESMTP id MAA19282 for ; Mon, 23 Oct 1995 12:15:02 -0700 Received: (from terry@localhost) by phaeton.artisoft.com (8.6.11/8.6.9) id LAA11337; Mon, 23 Oct 1995 11:17:36 -0700 From: Terry Lambert Message-Id: <199510231817.LAA11337@phaeton.artisoft.com> Subject: Re: NetBSD/FreeBSD (pthreads) To: cimaxp1!jb@werple.net.au (John Birrell) Date: Mon, 23 Oct 1995 11:17:36 -0700 (MST) Cc: leisner@sdsp.mc.xerox.com, hackers@FreeBSD.ORG, jb@cimlogic.com.au In-Reply-To: <199510202241.IAA26825@werple.net.au> from "John Birrell" at Oct 21, 95 08:44:09 am X-Mailer: ELM [version 2.4 PL24] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Length: 1392 Sender: owner-hackers@FreeBSD.ORG Precedence: bulk > > 3) Anyone have good hard numbers about the differences between user/kernel > > level threads on performance? > > No. We don't have a user-space implementation that works on a system with > kernel threads. We would have just used the kernel threads (like we do under > OSF/1^H^H^H^HDigital UNIX). 8-). I reimplemented Sun's LWP library, first on SunOS, then on BSD, and then later on Solaris and SVR4.2 (UnixWare). It's unfortunately one of the things that Novell felt "competed" with them (incorrectly: Novell didn't own USL at the time: USL felt they assumed all ownership, post-facto); I'd have to rewrite it to release it. The only performance comparisons I could make were on Solaris and SVR4.2. On the Solaris, there was a slightly higher system time and a lower user space time for kernel threads. On a Uniprocessor system, there was no significant difference. On SVR4.2 there was about a 12% advantage to user space threads on a UP system (probably attributable to their abominable page management practices compared to Sun). On MP systems, Solaris and SVR4 kernel threads both kicked user space threads butt's (but my test did not rely on a great deal of inter thread synchronization, so take it with a grain of salt). Regards, Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers.