From owner-freebsd-current@FreeBSD.ORG Fri Apr 22 23:07:04 2005 Return-Path: Delivered-To: freebsd-current@freebsd.org Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id EB66E16A4CE for ; Fri, 22 Apr 2005 23:07:04 +0000 (GMT) Received: from postal1.es.net (postal1.es.net [198.128.3.205]) by mx1.FreeBSD.org (Postfix) with ESMTP id B2EA143D41 for ; Fri, 22 Apr 2005 23:07:04 +0000 (GMT) (envelope-from oberman@es.net) Received: from ptavv.es.net ([198.128.4.29]) by postal1.es.net (Postal Node 1) with ESMTP (SSL) id IBA74465; Fri, 22 Apr 2005 16:07:04 -0700 Received: from ptavv (localhost [127.0.0.1]) by ptavv.es.net (Tachyon Server) with ESMTP id 1D9FB5D08; Fri, 22 Apr 2005 16:07:04 -0700 (PDT) To: "Daniel Eriksson" In-reply-to: Your message of "Fri, 22 Apr 2005 18:55:34 +0200." Date: Fri, 22 Apr 2005 16:07:04 -0700 From: "Kevin Oberman" Message-Id: <20050422230704.1D9FB5D08@ptavv.es.net> cc: 'FreeBSD Current' Subject: Re: Serious I/O problems (bad performance and live-lock) X-BeenThere: freebsd-current@freebsd.org X-Mailman-Version: 2.1.1 Precedence: list List-Id: Discussions about the use of FreeBSD-current List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 22 Apr 2005 23:07:05 -0000 > From: "Daniel Eriksson" > Date: Fri, 22 Apr 2005 18:55:34 +0200 > Sender: owner-freebsd-current@freebsd.org > > Scott Long wrote: > > > Ok, next test would be to turn off mpsafevm. This is just a simple > > guess, though, but and changes in the behaviour would be very > > interesting to note. > > With debug.mpsafevfs="0" and debug.mpsafevm="0" it doesn't seem like I can > force the machine into live-lock. No matter how long I let 'dd' run I can > still abort it with CTRL-C easily. However, after a few minutes the machine > becomes mostly useless because it takes 30-60 seconds to execute simple > commands such as ps or ls. > > According to 'top', the dd process spend most of its time in "wdrain" state. > > /Daniel Eriksson This certainly makes something to do with VM seem like the culprit, but I thought I'd post information on what I am seeing in hopes that it might help. I am still seeing the live-locks, too. They seem to happen when I have a CPU intensive process running. I have seen it with evolution, and transcode. The behavior is odd in that some things run fine. I can leave top running and it updates like normal and accepts commands like kill and renice. But I can get an interactive prompt either in X or on a vty. Suspending the CPU intensive process immediately frees up the live-lock. Running the CPU intensive job a nice 10 does not help. My Gnome applets are fine and seem to be very responsive. I have seen gkrellm lock up, but it usually does not. I am running ULE, so it's not 4BSD specific. I am also running PREEMPTION. System is an IBM T30 with 1.8 GHz CPU and .5 GB RAM. the kernel does not include SMP or APIC but does run ACPI. I am providing this in hopes of narrowing things down. The live-lock problem is not the cause of the I/O issues I was having with my fxp Ethernet or with my secondary IDE bus. -- R. Kevin Oberman, Network Engineer Energy Sciences Network (ESnet) Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab) E-mail: oberman@es.net Phone: +1 510 486-8634