From owner-freebsd-hackers Fri Oct 30 14:37:01 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id OAA04001 for freebsd-hackers-outgoing; Fri, 30 Oct 1998 14:37:01 -0800 (PST) (envelope-from owner-freebsd-hackers@FreeBSD.ORG) Received: from wrath.cs.utah.edu (wrath.cs.utah.edu [155.99.198.100]) by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id OAA03995 for ; Fri, 30 Oct 1998 14:36:58 -0800 (PST) (envelope-from danderse@cs.utah.edu) Received: from lal.cs.utah.edu (lal.cs.utah.edu [155.99.192.110]) by wrath.cs.utah.edu (8.8.8/8.8.8) with ESMTP id PAA29138; Fri, 30 Oct 1998 15:36:53 -0700 (MST) From: David G Andersen Received: (from danderse@localhost) by lal.cs.utah.edu (8.8.8/8.8.8) id PAA01262; Fri, 30 Oct 1998 15:37:02 -0700 (MST) Message-Id: <199810302237.PAA01262@lal.cs.utah.edu> Subject: nfs/amd hangs / getattr request flood problem To: hackers@FreeBSD.ORG Date: Fri, 30 Oct 1998 15:37:02 -0700 (MST) Cc: sclawson@cs.utah.edu, danderse@cs.utah.edu, mike@fast.cs.utah.edu X-Mailer: ELM [version 2.4 PL25] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-freebsd-hackers@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG We're in the process of configuring some new machines (for personal and distributed build farm use), and we're seeing some atrocities with amd. The machines are running 3.0-RELEASE (plus the last few days of checked in fixes). They receive AMD maps via NIS and a static map, but disabling NIS doesn't affect things. We've made significant tweaks to the rest of the system configuration (disabling nis, mfs, slowing things down, etc) and tried it on multiple systems, and the problem keeps popping up. This behavior isn't exhibited in 2.2.x. We have AMD looking at /n/{machine}/path, with the actual mounts on /a/{machine}. When compiling with a source tree on /n/machine/path and an object tree on local /z, AMD can use up to 50% of the processor. Ktrace and tcpdump output shows that it's handling around 150 getattr requests per second, on "/n" and "/n/machine", and the ktrace indicates that that's the _only_ thing it's doing. The result of this is some serious slowdowns, and reproducible system freezes (kernel alive and pingable, no userland activity whatsoever). Unfortunately, we don't have a simple way of reproducing the hangs yet - it happens while running the Linux suse Matrox Xserver and the FreeBSD netscape (entering, then leaving, and then re-entering Netscape frequently triggers the hang). There don't seem to be any references to this in gnats or on the lists. We're working on forward-porting the 2.2.x amd to 3.0 to see if the behavior still exists, but in the meantime, if anyone has suggestions / thoughts / knows what's wrong and wants to clue me in, it'd be greatly appreciated. :) -Dave -- work: danderse@cs.utah.edu me: angio@pobox.com University of Utah http://www.angio.net/ Department of Computer Science To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message