From owner-freebsd-smp Fri Apr 25 08:21:15 1997 Return-Path: Received: (from root@localhost) by hub.freebsd.org (8.8.5/8.8.5) id IAA22092 for smp-outgoing; Fri, 25 Apr 1997 08:21:15 -0700 (PDT) Received: from Ilsa.StevesCafe.com (Ilsa.StevesCafe.com [205.168.119.129]) by hub.freebsd.org (8.8.5/8.8.5) with ESMTP id IAA22077 for ; Fri, 25 Apr 1997 08:21:11 -0700 (PDT) Received: from Ilsa.StevesCafe.com (localhost [127.0.0.1]) by Ilsa.StevesCafe.com (8.8.5/8.8.5) with ESMTP id JAA07978; Fri, 25 Apr 1997 09:18:41 -0600 (MDT) Message-Id: <199704251518.JAA07978@Ilsa.StevesCafe.com> X-Mailer: exmh version 2.0gamma 1/27/96 From: Steve Passe To: NISHIO Shuichi cc: freebsd-smp@freebsd.org Subject: Re: Tyan ATX1668 problem -- success In-reply-to: Your message of "Fri, 25 Apr 1997 21:46:45 +0900." <19970425214645B.nishio@elysium.kecl.ntt.co.jp> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Date: Fri, 25 Apr 1997 09:18:41 -0600 Sender: owner-smp@freebsd.org X-Loop: FreeBSD.org Precedence: bulk Hi, > So, last Sunday, I cvsup'ed the SMP code and the entire current, did a > 'make World', and tried the SMP kernel with this fix and APIC_IO. > ... > However, there still was some problems. > > 1. /bin/ps -a sometimes freezes > ... > 2. panic on heavy traffic > ... > I also experienced crashes on dumping filesystems to DAT, but this > might be due to the problem with the AIC driver. > > Last evening, I tried a kernel with XFAST_IPI32 defined (and also the > latest current), and these problems seems to go away: > I tried > making kernel with -j8 > 'put'ting a 180MB file from FreeBSD to SGI O2 on 100BaseTX > 'get'ting a 60MB file from O2 to FreeBSD > dumping /home to DAT > > simultaneously, but it went through, without going panic. > > Additionaly, I don't know why, but the network transfer rate seems to > have raised to 55Mb/s after defining XFAST_IPI32, from 45Mb/s with the > previous kernel. > > So, in summary, the recent fix including XFAST_IPI32 seems to be > working very fine here. This is indeed good news! I would like to thank everyone who helped with this one, especially Cyrus, who spent many hours identifying the exact cause of the problem. I would also like anyone not running a 'critical' system to enable XFAST_IPI32 in smptest.h so I can hear that "it doesn't break anything here" a few times b4 making it mainline. As to the improved performance, this is because we are taking a big hit on all those TLB invalidations (has to be done), and the old method was competing for the kernel lock each time. The new method goes right for it without the lock. -- Steve Passe | powered by smp@csn.net | Symmetric MultiProcessor FreeBSD