home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.protocols.nfs
- Path: sparky!uunet!spool.mu.edu!sgiblab!sgigate!sgi!rhyolite!vjs
- From: vjs@rhyolite.wpd.sgi.com (Vernon Schryver)
- Subject: Re: NFS packet size
- Message-ID: <ugkusv0@rhyolite.wpd.sgi.com>
- Organization: Silicon Graphics, Inc. Mountain View, CA
- References: <u97b4fc@rhyolite.wpd.sgi.com> <C04t4o.K10@news.iastate.edu> <peterd.726170584@pjd.dev.cdx.mot.com>
- Date: Tue, 5 Jan 1993 19:27:53 GMT
- Lines: 45
-
- In article <peterd.726170584@pjd.dev.cdx.mot.com>, peterd@pjd.dev.cdx.mot.com (Peter Desnoyers) writes:
- > vjs@rhyolite.wpd.sgi.com (Vernon Schryver) writes:
- >
- > >In article <1993Jan1.115015.17311@prl.dec.com>, boyd@prl.dec.com (Boyd Roberts) writes:
- > >> ...
- > >> > A dropped TCP fragment can result in a large amount of resending too.
- > >>
- > >> True, but that all happens at the TCP/IP level which is far more
- > >> responsive than NFS's 1, 2, 4, 8 ... _second_ exponential backoff
- > >> when a datagram gets lost.
- >
- > >In other implementations of NFS, the NFS timeout starts at 0.7 seconds
- > >not 1.0, and can be reduced to 0.1 seconds. Please compare 0.1 seconds
- > >with the standard TCP timeouts.
- >
- > It still doesn't compare very well, at least on paper. With one second
- > timeouts and 8K blocks, exponential retransmission ala NFS falls off
- > to 50% throughput at an Ethernet packet loss rate of 0.001, and to 25%
- > throughput at a loss rate of 0.004. With a 100 mS starting timeout,
- > throughput falls to 50% at a loss rate of 0.01, and 25% at 0.025 or
- > so. (all figures from eyeballing gnuplot graphs...)
-
- Yes, 1.0 second NFS timeouts are painful. Even at 0.7 sec, a 1% packet
- loss produces a 90% NFS throughput drop, as I have observed from
- involuntary experiments.
-
- But we're not talking about 1.0 or even 0.7 sec NFS timeouts, but about
- 0.1 sec. NFS timeouts. I don't know how a 0.1 sec NFS retransmission
- timeout would work. It might be fine with few enough unneeded
- retransmissions, since any reasonable NFS server does far more than 10
- reads or writes per second.
-
-
- > With TCP on a local net, however, the retransmission timer should
- > adapt to somewhere in the range of say 3 to 6 packet times (correct me
- > if I'm wrong...), and so in the worst case (loss rate 0.025, timer = 6
- > packet times) the throughput would be about 87%.
- > ...
-
- In unmodified 4.3BSD style network code, the "fast" timer runs at 5 HZ,
- which means the smallest TCP timer and quickest retransmission you can
- have is 0.2 seconds.
-
-
- Vernon Schryver, vjs@sgi.com
-