home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.unix.ultrix
- Path: sparky!uunet!think.com!spool.mu.edu!sol.ctr.columbia.edu!destroyer!news.iastate.edu!john
- From: john@iastate.edu (John Hascall)
- Subject: Re: Slow writes to Ultrix 4.3 NFS server
- Message-ID: <BxsGJ1.A58@news.iastate.edu>
- Keywords: NFS server performance
- Sender: news@news.iastate.edu (USENET News System)
- Organization: Iowa State University, Ames, IA
- References: <leb.721782917@Hypatia> <BxqyCu.791@news.iastate.edu> <sdlhiig@zuni.esd.sgi.com>
- Date: Mon, 16 Nov 1992 03:04:12 GMT
- Lines: 60
-
- attribution-long-gone writes:
- }}>}I manage an SGI 4D/340 running IRIX 4.0.1 and a DECstation 5000/125 running
- }}>}Ultrix 4.3.
-
- john@iastate.edu (John Hascall) writes:
- >Anyway, several vendors' NFS servers mis-implement the standard (on purpose).
- >The NFS standard requires that writes be done "synchronously".
-
- olson@anchor.esd.sgi.com (Dave Olson) writes:
- }Well, not actually. It requires that it be written to 'stable storage'.
- }There have been many hundreds of hours spent on arguing this point. SGI
- }feels that our uptime is good enough that by the time it makes it to RAM
- }on the server, it is in 'stable storage'. We also provide an option
- }to do the 'synchronous to disk' version, which many people prefer.
-
- Here's what the standard [RFC1094] says:
- All of the procedures in the NFS protocol are assumed to be synchronous. When
- a procedure returns to the client, the client can assume that the operation
- has completed and any data associated with the request is now on stable
- storage. For example, a client WRITE request may cause the server to update
- data blocks, filesystem information blocks (such as indirect blocks), and file
- attribute information (size and modify times). When the WRITE returns to the
- client, it can assume that the write is safe, even in case of a server crash...
- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
- }No worse than writes on your local disk, since they are of course
- }buffered also.
-
- Yes, but you have a pretty good clue when your local machine crashes, it is
- possible to have a server reboot and the for the client to not know this.
-
- }There is no 'standard' in this sense for NFS, only implementations, or
- }such is my understanding from all the arguments/discussions.
-
- RFC1094 NFS: Network File System Protocol specification. Sun Microsystems, Inc.
- 1989 March; 27 p.
-
- }Again, no difference between NFS and local disk, in this case, so why make
- }such a big deal about it. The only case I can see any justification for
- }the argument is if you have a system set up purely as an NFS server, similar
- }to Auspex (that is, where you don't have any local disk writes to speak of).
- }In that case, I would definitely expect a UPS, or sync writes. However,
- }you explictly mention workstations, and for workstations, you will have
- }far more local writes than you will NFS writes, so the argument seems rather
- }silly.
-
- I don't think this is universally true by any means. For example, we have
- a number of SGI labs (not setup by me!) on campus with N machines (N typically
- 15-25) which all mount each other's disks for home directories, so there is
- only a 1/Nth chance your home dir is local.
-
- I am, of course, playing the devil's advocate here -- we do run asynch writes
- on our NFS servers -- but, at least we went into it with our eyes wide open.
-
- John
- --
- John Hascall ``An ill-chosen word is the fool's messenger.''
- Systems Software Engineer
- Project Vincent
- Iowa State University Computation Center + Ames, IA 50011 + 515/294-9551
-