Date: Wed, 16 Dec 1998 01:44:00 +0100 From: Bernd Walter <ticso@cicely.de> To: Charles Owens <owensc@enc.edu>, hackers@FreeBSD.ORG Subject: Re: NFS thoughts Message-ID: <19981216014400.20185@cicely.de> In-Reply-To: <Pine.BSF.3.96.981215092411.2552B-100000@itsdsv2.enc.edu>; from Charles Owens on Tue, Dec 15, 1998 at 10:10:16AM -0500 References: <Pine.BSF.3.96.981215092411.2552B-100000@itsdsv2.enc.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Dec 15, 1998 at 10:10:16AM -0500, Charles Owens wrote: > On Mon, 14 Dec 1998, Bernd Walter wrote: > > > I saw the same on my private hosts. > > Everythings the same to your case instead that I have a 100MBit FreeBSD Router > > between them. All lines are running Full-Duplex Point-to-Point. > > In my case I have a syslogentry telling me about a server down under some load > > and it took minutes till it says that the server is up again. > > It happend when using NFS3/TCP at this moment I'm using NFS2/UDP and it won't > > hang. > > I was at the Usenix LISA conference last week and in one of the > tutorials the issue of NFSv3/TCP stability and interoperability came > up. > > One user reported the SGI<->Solaris and even SGI<->SGI NFSv3 mounts > were flakey (with similar symptoms as described here). He eventually > traced the problem to the NFSv3/TCP's use of larger buffersizes. > These result in more intense bursts of network activity which would at > times overrun buffers in his Ethernet switch. He convinced his A switch should never have overruns. If a buffer gets full a switch is able to top receiving. On Half-Duplex lines a switch usualy create some collinsions on the receiving port top slow down thne sender. On Full-Duplex Ports there's a compareable method for doing the same. In any case it's a broken switch, nic, nic-driver or configuration of them. Very often people don't check their ethernets for errors due to Duplex-missonfigurations > network vendor to replace the switch with another with deeper buffers, > and his problem went away! > > With the original switch he found that the default 32K read/write > buffer size (as compared to UDP's default of 8K) was too much, but by > limiting it to 16K he was able to get by, but with some reduction in > performance. > > Could this be the issue that is plaguing our attempts to use NFSv3 > with FreeBSD? Or are there other known defects that are at fault. In my case there is only a FreeBSD Router between them. All Line are without any known problems. > > I'm guessing that with FreeBSD we'd use the -r and -w options of mount_nfs > to limit the read and write buffer sizes, though the manpage description > isn't quite as informative as what's provided with the Solaris: > > rsize=n Set the read buffer size to n bytes. > The default value is 32768 when using > Version 3 of the NFS protocol. When > using Version 2, the default value is > 8192. > > Does this mount_nfs option from Solaris indeed work the same way as > FreeBSD's -r option? > > later, > --- > ------------------------------------------------------------------------- > Charles N. Owens Email: owensc@enc.edu > http://www.enc.edu/~owensc > Network & Systems Administrator > Information Technology Services "Outside of a dog, a book is a man's > Eastern Nazarene College best friend. Inside of a dog it's > too dark to read." - Groucho Marx > ------------------------------------------------------------------------- > > > > > To Unsubscribe: send mail to majordomo@FreeBSD.org > with "unsubscribe freebsd-hackers" in the body of the message -- B.Walter To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19981216014400.20185>