From owner-freebsd-smp Wed Jun 13 13: 9:36 2001 Delivered-To: freebsd-smp@freebsd.org Received: from pynchon.icsi.berkeley.edu (katje.aciri.org [192.150.187.51]) by hub.freebsd.org (Postfix) with ESMTP id BB63737B40A for ; Wed, 13 Jun 2001 13:09:33 -0700 (PDT) (envelope-from bkarp@icsi.berkeley.edu) Received: (from karp@localhost) by pynchon.icsi.berkeley.edu (8.11.3/8.11.3) id f5DK41a00672 for freebsd-smp@freebsd.org; Wed, 13 Jun 2001 13:04:01 -0700 (PDT) (envelope-from karp) Date: Wed, 13 Jun 2001 13:04:01 -0700 (PDT) From: Brad Karp Message-Id: <200106132004.f5DK41a00672@pynchon.icsi.berkeley.edu> To: freebsd-smp@freebsd.org Subject: clock speedup on 4.3 SMP boxes Sender: owner-freebsd-smp@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.org I've been experiencing clock speedup (that is, FreeBSD's notion of the current time of day advances faster than real time) on two SMP boxes that run 4.3-RELEASE. The two motherboards: - Tyan S2567, with ServerWorks HEsl chipset - Intel STL2 On both motherboards, I use dual 1 GHz P-IIIs. The clock acceleration appears to be directly proportional to the CPU load on the machine. When I run large simulations that drive one or both CPUs to 100% utilization, the clock gain approaches one hour gain (or more!) per 24 hours real time! Though I run ntpd, ntpd quits once the clock drifts so far so fast, of course. I scoured the -smp archives, assuming that I'd see other reports if this is a kernel SMP code bug. But I found no such reports. At that point, I'd only seen the behavior on the Tyan S2567. So I sent the board to Tyan for repair! Then, I began observing the same behavior on the Intel STL2 board. I now fear that this is an SMP kernel bug. Has anyone seen such clock gains under 4.3, or earlier releases? Many thanks, -Brad, bkarp@icsi.berkeley.edu To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-smp" in the body of the message