From owner-freebsd-stable Thu Mar 22 8:31: 8 2001 Delivered-To: freebsd-stable@freebsd.org Received: from dirac.th.physik.uni-bonn.de (dirac.th.physik.uni-bonn.de [131.220.161.119]) by hub.freebsd.org (Postfix) with SMTP id 70D7E37B71A for ; Thu, 22 Mar 2001 08:31:01 -0800 (PST) (envelope-from conrad@th.physik.uni-bonn.de) Received: (qmail 44980 invoked from network); 22 Mar 2001 16:30:59 -0000 Received: from merlin.th.physik.uni-bonn.de (131.220.161.121) by dirac.th.physik.uni-bonn.de with SMTP; 22 Mar 2001 16:30:59 -0000 Received: (qmail 1043 invoked by uid 145); 22 Mar 2001 16:30:59 -0000 Received: from localhost (sendmail-bs@127.0.0.1) by localhost with SMTP; 22 Mar 2001 16:30:59 -0000 Date: Thu, 22 Mar 2001 17:30:58 +0100 (CET) From: Jan Conrad To: Matt Dillon Cc: Helge Oldach , , , , Subject: Re: NFS performance In-Reply-To: <200103212103.f2LL3h420596@earth.backplane.com> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-freebsd-stable@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG On Wed, 21 Mar 2001, Matt Dillon wrote: >.... > :again, running on half-duplex, transfering 100Mb from a client to this > :machine (merlin) > :on client: > :mount -t nfs -o intr,nfsv3,-r=32768,-w=32768 merlin:/freebsd/misc /mnt > :dd if=/dev/zero of=/mnt/zero bs=16k count=64x100 > :104857600 bytes transferred in 12.765062 secs (8214422 bytes/sec) > : > :at the same time, netstat -I fxp0 -w 1 gives me > > Those numbers are fairly good for a half-duplex link. It should work > almost as well with the default blocksize of 8K, even with the additional > return traffic. If it doesn't, something is wrong somewhere. Well, see below. Any hints on how to find out what might be wrong? > > In anycase, setting the interface and the switch to full duplex > (if possible) would give you even better results. You should be > able to max-out the transfer rate for writes at 10MB/sec or so even > using the default 8K block size. Hmm, today we set our two boxes to full-duplex. With UDP, block size=8192 we get around 3Mb/s (tcp 3.5Mb/s) With UDP, block size=16384 we get around 9.5 Mb/s (tcp 9Mb/s) With UDP, block size=32768 we get around 9.3 Mb/s (tcp 5.4Mb/s) > > Ultimately these sequential transfer numbers are meaningless. When > dealing with heavier loads from multiple clients the default block > size of 8K should typically work better. > > -Matt > -- Physikalisches Institut der Universitaet Bonn Nussallee 12 D-53115 Bonn GERMANY To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message