From: Greg Banks Subject: Re: [PATCH] NFS: Zeroing NFS and kNFSD stats Date: Thu, 5 Aug 2004 15:26:19 +1000 Sender: nfs-admin@lists.sourceforge.net Message-ID: <20040805052619.GT5581@sgi.com> References: <40F3D483.8020507@RedHat.com> <20040713143044.GA3023@fieldses.org> <40F3FAD8.7090707@RedHat.com> <20040713151759.GC3023@fieldses.org> <20040802103905.GI25023@suse.de> <20040804201411.GL19282@fieldses.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Olaf Kirch , Steve Dickson , Linux NFS Mailing List Return-path: Received: from sc8-sf-mx1-b.sourceforge.net ([10.3.1.11] helo=sc8-sf-mx1.sourceforge.net) by sc8-sf-list2.sourceforge.net with esmtp (Exim 4.30) id 1Bsaly-0000Ge-KU for nfs@lists.sourceforge.net; Wed, 04 Aug 2004 22:26:38 -0700 Received: from omx3-ext.sgi.com ([192.48.171.20] helo=omx3.sgi.com) by sc8-sf-mx1.sourceforge.net with esmtp (Exim 4.34) id 1Bsalx-0002Zn-Al for nfs@lists.sourceforge.net; Wed, 04 Aug 2004 22:26:37 -0700 To: "J. Bruce Fields" In-Reply-To: <20040804201411.GL19282@fieldses.org> Errors-To: nfs-admin@lists.sourceforge.net List-Unsubscribe: , List-Id: Discussion of NFS under Linux development, interoperability, and testing. List-Post: List-Help: List-Subscribe: , List-Archive: On Wed, Aug 04, 2004 at 04:14:11PM -0400, J. Bruce Fields wrote: > On Mon, Aug 02, 2004 at 12:39:05PM +0200, Olaf Kirch wrote: > > On Tue, Jul 13, 2004 at 11:17:59AM -0400, J. Bruce Fields wrote: > > But it sucks. I had folks run something like 12-15 performance > > measurements and take the NFS stats, and it was a huge mess to sort > > through. Telling folks "use nfsstat -z to wipe stats before each > > command" is much easier. > > Well, like I say, we could implement "nfsstat -z" by checkpointing as > above but just writing to a file at some default path, though admittedly > the choice of path might not be obvious. FWIW, futzing with zeroing or checkpointing should not be necessary on Altix systems because they ship with PCP, a stats gathering package which samples the NFS stats (among many others) and does rate conversion. http://oss.sgi.com/projects/pcp/ > Having nfsstat just hang around till it's signaled (like tcpdump does) > would also be easy. IRIX' nfsstat has a -C option which uses curses to display every 1 sec the rates of various NFS stats, until interrupted. It's a bit like "watch nfsstat" but shows rate converted numbers instead of changing counters. Here's an example screen, for a machine getting 30 RPC calls/sec: 4: Server snort Aug 5 15:21:21 D: Delta/second Server RPC: calls badcalls nullrecv badlen xdrcall duphits dupage 30 0 0 0 0 0 40086.17 Server NFS V3: calls badcalls 30 0 null getattr setattr lookup access readlink 1 0% 10 33% 1 2% 16 52% 1 0% 0 0% read write create mkdir symlink mknod 2 4% 1 3% 0 0% 0 0% 0 0% 0 0% remove rmdir rename link readdir readdir+ 0 0% 0 0% 0 0% 0 0% 0 0% 0 0% fsstat fsinfo pathconf commit 1 1% 0 0% 0 0% 1 2% 1: Client[V2] 2: Server[V2] 3: Client[V3] 4: Server[V3] 5: Daemons DZR:mode Something like that might be more useful than the -z option. Greg. -- Greg Banks, R&D Software Engineer, SGI Australian Software Group. I don't speak for SGI. ------------------------------------------------------- This SF.Net email is sponsored by OSTG. Have you noticed the changes on Linux.com, ITManagersJournal and NewsForge in the past few weeks? Now, one more big change to announce. We are now OSTG- Open Source Technology Group. Come see the changes on the new OSTG site. www.ostg.com _______________________________________________ NFS maillist - NFS@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/nfs