From: Orion Poplawski Subject: Re: Client performance questions Date: Thu, 13 Dec 2007 21:24:18 -0700 Message-ID: <47620572.4020501@cora.nwra.com> References: <0a15723c4b267d4eb8b5ad05800315c0@swip.net> <1197328226.25938.7.camel@heimdal.trondhjem.org> <1197593822.7964.8.camel@heimdal.trondhjem.org> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Cc: linux-nfs@vger.kernel.org To: Trond Myklebust Return-path: Received: from earth.cora.nwra.com ([65.44.101.180]:52606 "EHLO earth.cora.nwra.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755694AbXLNEeF (ORCPT ); Thu, 13 Dec 2007 23:34:05 -0500 In-Reply-To: <1197593822.7964.8.camel-rJ7iovZKK19ZJLDQqaL3InhyD016LWXt@public.gmane.org> Sender: linux-nfs-owner@vger.kernel.org List-ID: Trond Myklebust wrote: > On Thu, 2007-12-13 at 17:07 -0700, Orion Poplawski wrote: >> Trond Myklebust wrote: >>> Have you tried increasing the value >>> in /proc/sys/fs/nfs/nfs_congestion_kb ? >> Well, I just came across this message while trying to improve NFS >> performance locally. On my F8 client, it is set to 30400 by default. >> Changing to 60800 kills performance. Writes go from ~20MB/s to ~5MB/s, >> reads from ~80MB/s to ~47MB/s as measured by bonnie++. >> >> Expected? > > UDP? If so, then possibly yes. UDP has lousy congestion control, so if > the server gets too busy to respond, then performance dies. Nope, TCP. w/rsize is 32k. Server is pretty lightly loaded. Server is RHEL4.5. -- Orion Poplawski Technical Manager 303-415-9701 x222 NWRA/CoRA Division FAX: 303-415-9702 3380 Mitchell Lane orion@cora.nwra.com Boulder, CO 80301 http://www.cora.nwra.com