home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.protocols.nfs
- Path: sparky!uunet!sun-barr!ames!sgi!rhyolite!vjs
- From: vjs@rhyolite.wpd.sgi.com (Vernon Schryver)
- Subject: Re: NFS / RFS performance ???
- Message-ID: <pk7ckic@rhyolite.wpd.sgi.com>
- Organization: Silicon Graphics, Inc. Mountain View, CA
- References: <1992Sep8.192211.8020@vector.dallas.tx.us>
- Date: Wed, 9 Sep 1992 03:51:04 GMT
- Lines: 50
-
- In article <1992Sep8.192211.8020@vector.dallas.tx.us>, tbo@vector.dallas.tx.us (Terry Bohaning) writes:
- >
- > Does anyone have any experience with Sun RFS and NFS.
- >
- > We're currently using NFS in a major way, but have learned the hard way
- > what an "Unreliable Transport" that UDP is. From Sun's documentation it
- > appears that RFS is TCP based, but it really doesn't give me any idea
- > of the performance characteristics of the protocols. Can anyone give me
- > a comparison of the NFS vs RFS.
- >
- > I'm trying to work around the current nightmare that we have. Most of
- > our user's have quit using NFS as it is so unreliable. In order to get
- > enough speed, we need to run with a frame size of 8k. For reliability,
- > OTOH, we need to set the frame size between 1-2k. This means that we take
- > a hugh hit in performance to get a reliable transport.
- >
- > Will RFS give us a reasonably fast, reliable transport mechanism???
-
-
- Changing the packet size to 1K does not affect the reliability of
- NFS on competently designed and implemented NFS servers and clients.
- It will reduce performance somewhat.
-
- One needs small UDP packets only if:
- -using cheap (not necessarily inexpensive) ethernet hardware
- or operating systems on clients or servers that cannot handle
- several back-to-back ethernet packets.
-
- -using broken routers or bridges.
-
- -you have a badly broken ethernet, which loses a significant
- fraction (>= 1%) of its packets.
-
- Any of these situations will terrible things to any network protocol,
- albeit NFS is particularly vulnerable.
-
-
- Please note that NFS can be made entirely reliable, although
- arbitrarily slow, by using "hard mounts," even in such broken
- environments.
-
- Perhaps the most common NFS configuration error is to use soft mounts
- inappropriately (i.e. where reliability matters, such as on writable
- file systems).
-
-
- Vernon Schryver, vjs@sgi.com
-
- P.S. The "Distribution: usa" line in your article may have reduced
- the number of people who saw it.
-