Re: NFS reliability?

From: sinister (sinister_at_nospam.invalid)
Date: 11/11/04

  • Next message: sinister: "Re: NFS reliability?" 
    Date: Thu, 11 Nov 2004 12:35:08 GMT

    "Juhan Leemet" <juhan@logicognosis.com> wrote in message
    news:pan.2004.11.11.08.57.55.232387@logicognosis.com...
    > On Wed, 10 Nov 2004 12:49:37 +0000, Casper H. S. *** wrote:
    >> "sinister" <sinister@nospam.invalid> writes:
    >>>How reliable is NFS?
    >>
    >> About as reliable as the transport you use (so you get the usual
    >> ethernet CRC and UDP or TCP checksum protection). This is usually
    >> sufficient but those checksums do not protect against all errors;
    >> the best thing to do is use a flavour of secure RPC with integrity
    >> enabled. That will give you a secure checksum with a negligible
    >> chance of file corruption. (Data corruption which passes the TCP
    >> and UDP checksums does occur)
    >>
    >>>That is, if a directory subtree (for example) is mv'ed across NFS, what's
    >>>the likelihood that the data are not 100% preserved?
    >>
    >> Small but it's hard to put a finger on.
    >
    > FWIW, and I will admit that this is "anecdotal" (i.e. not any kind of
    > documented reproducible test case), but it might depend on the client
    > and/or server platform. I could swear that I have had "stuff evaporate"
    > when working on (hated word?) Linux NFS clients, accessing files on
    > Solaris8 or 9 NFS servers. Typically this happens when renaming files in
    > script loops, with overlaps between old and new filename ranges. I have
    > gone back to carefully examine shell history and the files that were
    > supposed to be there were gone. I had backups, but it was unsettling.
    > It's almost as though some of the directory operations got out of sequence
    > (dunno the internal details). As I say, this is just anecdotal, but maybe
    > something is still busted in Linux NFS. Be careful out there. YMMV
    >
    > You don't say, but since 2 of the newsgroups you posted to are sun and
    > solaris, can we assume that both client and server are Solaris machines? I

    Right. Two SunOs 5.9 machines.

    > would have more confidence in that combination. The Linux client usually
    > works, too, but there have been 1 or 2 times over the last number of
    > years...
    >
    >>>Furthermore, does NFS include checksums to ascertain the fidelity of the
    >>>transfer?
    >>
    >> See above.
    >>
    >> Casper
    >
    > --
    > Juhan Leemet
    > Logicognosis, Inc.
    >

  • Next message: sinister: "Re: NFS reliability?"
    Loading