Article View: comp.dcom.lans.fddi
Article #330Re: NFS over FDDI considerations
From: Robert Macomber
Date: Mon, 04 Jan 1993 14:12
Date: Mon, 04 Jan 1993 14:12
45 lines
2725 bytes
2725 bytes
In article <1992Dec30.050448.21737@grebyn.com>, rich@grebyn.com (Richard
Lawrence) writes:
>
> Forgive me if this is a FAQ.
>
> I seem to remember coming across a document detailing special
> considerations when running NFS over FDDI. I can't remember the
> specifics, but there was an optimal and inefficient way to do it, and
> the article implied that many people were doing it the inefficient way.
>
> The only thing I can think of is configuring the MTU for better
> throughput with NFS, but I can't think of why this would have a
> significant effect...anyone? Offhand I can't remember if Sun decided 4K
> or 8K packets were the sanctioned unit for NFS..probably 8K, but I'm
> guessing there.
>
The IP MTU for FDDI is typically set at 4352 bytes, but I don't think that
this what the document is talking about (I haven't seen it, though). The
MTU is usually established by the device driver (although on
AIX 3.2 it may be dynamically adjusted. The document is probably talking
about increasing the size of UDP datagrams; which may be sent
via one or more IP packets. There are two things (that I know of)
that control this aspect of NFS operation: NFS rsize/wsize settings
(in /etc/fstab, /etc/filesystems. or whatever)
and udp_sendspace/udp_recvspace.
SunOS (4.1) is shipped with an "in_proto.c" file that configures the udp/tcp
send/recv spaces to sizes that apparently work well with Ethernet. This
file can be found at /usr/sys/netinet/in_proto.c. The maximum datagram size
and maximum TCP window size is constrained by the values in this file. On
AIX 3.2 the "no" command can be used to dynamically alter these values.
I think that SunOS uses a default configuration of wsize=8K and rsize=8K.
Does anyone know where this document can be found? It isn't clear to me that
increasing the datagram size will have a uniformly positive effect (though,
fewer NFS write syncs would be nice).
- Robert
Robert L. Macomber MS 11-21 Advanced Communications
National Semiconductor Corp. Internet: macomber@thoreau.NSC.COM
333 Western Avenue UUCP: {sun|hplabs|decwrl}!nsc!thoreau!macomber
South Portland, Maine 04106 My opinions are my mine.
Technical Details
Message-ID:
<1993Jan4.141212.9943@berlioz.nsc.com>Path:
arc1-ca.pugleaf.net!sparky!uunet!psinntp!gatekeeper.nsc.com!voder!berlioz.nsc.com!thoreau.nsc.com!macomberReferences:
<1992Dec30.050448.21737@grebyn.com>Show Full Headers
Complete Article Headers
Newsgroups: comp.dcom.lans.fddi Path: arc1-ca.pugleaf.net!sparky!uunet!psinntp!gatekeeper.nsc.com!voder!berlioz.nsc.com!thoreau.nsc.com!macomber From: macomber@thoreau.nsc.com ( Robert Macomber) Subject: Re: NFS over FDDI considerations Message-ID: <1993Jan4.141212.9943@berlioz.nsc.com> Sender: news@berlioz.nsc.com (UseNet News account) Reply-To: macomber@thoreau.nsc.com ( Robert Macomber) Organization: Nat'l Semi References: <1992Dec30.050448.21737@grebyn.com> Date: Mon, 4 Jan 1993 14:12:12 GMT Lines: 45