From owner-freebsd-questions Tue Jun 4 9:31: 4 2002 Delivered-To: freebsd-questions@freebsd.org Received: from tltodd.com (badger.tltodd.com [208.133.92.209]) by hub.freebsd.org (Postfix) with ESMTP id B021F37B405 for ; Tue, 4 Jun 2002 09:30:58 -0700 (PDT) Received: (from tlt@localhost) by tltodd.com (8.9.3/8.9.3) id LAA28418 for freebsd-questions@freebsd.org; Tue, 4 Jun 2002 11:30:56 -0500 (CDT) (envelope-from tlt) Date: Tue, 4 Jun 2002 11:30:56 -0500 From: Terry Todd To: freebsd-questions@freebsd.org Subject: Re: FreeBSD 5.0dp1 nfs server to Red Hat nfs client extremely slow Message-ID: <20020604113056.A27948@badger.tltodd.com> References: <20020603102416.A98405@badger.tltodd.com> <20020603152547.A32050@xor.obsecurity.org> <20020604103039.A11587@badger.tltodd.com> <20020604155426.GB46396@dan.emsphone.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Mailer: Mutt 1.0.1i In-Reply-To: <20020604155426.GB46396@dan.emsphone.com>; from dnelson@allantgroup.com on Tue, Jun 04, 2002 at 10:54:26AM -0500 Sender: owner-freebsd-questions@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.ORG On Tue, Jun 04, 2002 at 10:54:26AM -0500, Dan Nelson wrote: > In the last episode (Jun 04), Terry Todd said: > > Here's more data: > > All comparisons were done by running: > > time dd if=/dev/zero of=somefile bs=1024k count=256 > > > > NFS server NFS client transfer time > > Red Hat 2.2.16-22 Slackware 2.2.20 0m25.48s > > Red Hat 2.2.16-22 FreeBSD 4.4-RELEASE 46.902824 secs > > Slackware 2.2.20 FreeBSD 4.4-RELEASE 49.100937 secs > > Slackware 2.2.20 Red Hat 2.2.16-22 0m40.375s > > FreeBSD 5.0-DP1 FreeBSD 4.4-RELEASE 52.279700 secs > > FreeBSD 5.0-DP1 Slackware 2.2.20 0m55.270s > > FreeBSD 5.0-DP1 Red Hat 2.2.16-22 18m12.448s <- 18 Minutes > > > > I can ftp 268435456 bytes from the FreeBSD 5.0-DP1 to > > the Red Hat 2.2.16-22 in 28 seconds. > > I assume the RedHat 2.2.16 kernel is using NFSv2? NFSv2 requires that > all writes be immediately commited to stable storage (i.e. either disk > or battery-backed RAM). Chances are the Linux NFS servers are caching > writes against spec, which is why they are so fast. A way to check > this is to just listen to the disks when you do the copy. An NFSv2 > server with no battery-backed RAM should be hammering its disks syncing > data (usually a seek per sync). > > If you want fast NFSv2 mounts, sysctl vfs.nfs.async=1 on the server, > but be aware that if the server crashes it could corrupt files in use > by the client. If you want fast reliable mounts, upgrade that Redhat > box to something that can do NFSv3. Dan, Thanks a million. vfs.nfs.async=1 was not exactly it but allowed me to grep the output of sysctl -a for async and come up with vfs.nfsrv.async which I set to 1 and viola my nfs write times from the RH system to the 5.0DP1 system went to 0m54.747s which is well within the acceptable range. I understand the rest of the implications. Thanks again, Terry Todd To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message