Date: Thu, 22 Mar 2001 12:05:33 +0100 (CET) From: "Hartmann, O." <ohartman@klima.physik.uni-mainz.de> To: Hurf Sheldon <hurf@Graphics.Cornell.EDU> Cc: Steve Curry <scurry505@yahoo.com>, <freebsd-questions@FreeBSD.ORG> Subject: Re: NFS performance tweaking Message-ID: <Pine.BSF.4.33.0103221132380.3086-100000@klima.physik.uni-mainz.de> In-Reply-To: <3AB918B5.A4F512A5@graphics.cornell.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, 21 Mar 2001, Hurf Sheldon wrote: Dear Sirs. All right. I got many responses and this message from Mr. Curry aimed aomething I use in the kernel: ICMP_BANDLIM. I set this parameter and I set via /etc/sysctl.conf a limitation of 256. My NMBCLUSTERS are at 4096 as a long time standard. I changed the icmp limitation from 256 to 1024 and will wait for results ... The other point of view targets the NIC. Our mainserver has a built-in NIC, it is a TYAN Thunder 2500 mainboard with a built in Intel/DEC chip EtherExpress Pro (83C559 chipset) without FlashROM. The Switch at which four systems are attached to is Intel 410T Switch. All of our machines running FreeBSD use the Intel EtherExpress 10/100PRO Adapter, and in the past performance seems to be very, very good! If there is any suggestion for a high performance 64Bit -PCI NIC (are their supported Intel 1GBit NICs with 10/100 MBit facilities for migration?)? Thanks a lot ... :>Assuming you don't have a network or card :>problem ( a quick look with netstat or at switch statistics for :>collision counts on affected systems can indicate a problem) :>Try this: :>Run `dmesg` and see if you have any messages :>from nfsgetpages or re: nmbuf(?) - if so (or in any case), :>upping "MAXFILES" and "NMBCLUSTERS" :>in the kernel may fix the problem. You can also do :>both by increasing "MAXUSERS", which is used :>as a seed in allocation macros. :> :>If you have a dmesg entry "ICMP_BANDLIM " :>you may want to take this out of the kernel and see if :>the problem goes away - we had a similar problem with a 3.2 :>based NFS server and after some digging around decided (for :>a good reason not recalled) that it was best on a heavily used NFS :>server :>to not have "ICMP_BANDLIM " in the kernel. :> :>(I'm assuming that you are familiar with building a kernel. If not :>it is easy to learn either from the handbook or the web pages :>or I'd be happy to send you pointers [which is has it's own :>risks ] ) :>hurf :> :> :>Steve Curry wrote: :> :>> Hartmann, :>> :>> It could be your timeout settings, but then again I :>> don't *really* know whats going on at your site (maybe :>> network issues?). I would recommend the "Managing NFS :>> and NIS" from O'Reilly. It was published back in 1991 :>> but it's still jam packed with good information. It :>> even has a chapter on checking performance and :>> tweaking, so it sounds as if it would be a good fit :>> for you. :>> :>> My 2 cents, :>> :>> Steve Curry :>> Technical Yahoo :>> Yahoo, Inc. :>> :>> --- "Hartmann, O." :>> <ohartman@klima.physik.uni-mainz.de> wrote: :>> > :>> > Dear Sirs. :>> > :>> > Our main NFS server fails very often in serving fast :>> > enough its :>> > services and on several clients I get the message :>> > "nfs server not :>> > responding", later "... live again". :>> > :>> > Sometimes the server is really under heavy loead, :>> > sometimes not. :>> > I think it is a task of "tweaking", I saw a lot of :>> > stuff in the kernel :>> > and via sysctl -A in the kernel paramameters. :>> > :>> > Does anyone has a suitable hint waht to set for :>> > these parameters :>> > (prefereable in the kernel ...)? :>> > :>> > Thanks a lot ... :>> > :>> > Oliver :>> > :>> > -- :>> > MfG :>> > O. Hartmann :>> > :>> > ohartman@klima.physik.uni-mainz.de :>> > :>> ---------------------------------------------------------------- :>> > IT-Administration des Institut fuer Physik der :>> > Atmosphaere (IPA) :>> > :>> ---------------------------------------------------------------- :>> > Johannes Gutenberg Universitaet Mainz :>> > Becherweg 21 :>> > 55099 Mainz :>> > :>> > Tel: +496131/3924662 (Maschinensaal) :>> > Tel: +496131/3924144 :>> > FAX: +496131/3923532 :>> > :>> > -- :>> > MfG :>> > O. Hartmann :>> > :>> > ohartman@klima.physik.uni-mainz.de :>> > :>> ---------------------------------------------------------------- :>> > IT-Administration des Institut fuer Physik der :>> > Atmosphaere (IPA) :>> > :>> ---------------------------------------------------------------- :>> > Johannes Gutenberg Universitaet Mainz :>> > Becherweg 21 :>> > 55099 Mainz :>> > :>> > Tel: +496131/3924662 (Maschinensaal) :>> > Tel: +496131/3924144 :>> > FAX: +496131/3923532 :>> > :>> > :>> > To Unsubscribe: send mail to majordomo@FreeBSD.org :>> > with "unsubscribe freebsd-questions" in the body of :>> > the message :>> > :>> > :>> > :>> > :>> :>> __________________________________________________ :>> Do You Yahoo!? :>> Get email at your own domain with Yahoo! Mail. :>> http://personal.mail.yahoo.com/ :>> :>> To Unsubscribe: send mail to majordomo@FreeBSD.org :>> with "unsubscribe freebsd-questions" in the body of the message :> :>-- :>Hurf Sheldon :>Dir. Research Systems :>Program of Computer Graphics :>580 Rhodes Hall, Hoy Rd. :>Cornell University :>Ithaca, N.Y. 14853 :> :>voice:607 255 6713 fax:607 255 0806 :>email: hurf@graphics.cornell.edu :>http://www.graphics.cornell.edu/~hurf/ :> :> :> -- MfG O. Hartmann ohartman@klima.physik.uni-mainz.de ---------------------------------------------------------------- IT-Administration des Institut fuer Physik der Atmosphaere (IPA) ---------------------------------------------------------------- Johannes Gutenberg Universitaet Mainz Becherweg 21 55099 Mainz Tel: +496131/3924662 (Maschinensaal) Tel: +496131/3924144 FAX: +496131/3923532 To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.33.0103221132380.3086-100000>