ver3 not being used - NFS

This is a discussion on ver3 not being used - NFS ; We have a java application on SunOS 5.8 that is attempting access an automounted (amd) NFS resource. I am guessing amd is taking too long to mount the resource, so the application gives up. Since this is SunOS 5.8 the ...

+ Reply to Thread
Results 1 to 4 of 4

Thread: ver3 not being used

  1. ver3 not being used

    We have a java application on SunOS 5.8 that is attempting access an
    automounted (amd) NFS resource. I am guessing amd is taking too long
    to mount the resource, so the application gives up. Since this is
    SunOS 5.8 the default mount option is NFSv3/TCP, and ‘nfsstat –m' is
    telling me that all NFS mounts are using v3/TCP . ‘nfsstat' is also
    telling me that there were 0 "connection oriented" timeouts, but that
    there were a few "connectionless" timeouts. Confusingly, ‘nfsstat'
    says that just over 50% of total nfs calls are v2, all of which are
    lookup, readdir and readlink. How can over 50% of NFS calls be v2,
    when everything that is mounted using v3?

    I looked at a few other machines in our environment and it looks like
    all of them are doing the same thing (all v3 mounts, but just over 50%
    of calls are v2). I looked at the NFS server and it has 1020 UDP
    calls from the server in question but not one NFS connectionless
    operation. Why are there so many UDP calls that appear to just be
    asking for a directory listing, and how/should I stop it from doing
    that?

    Thanks,
    Jeff

  2. Re: ver3 not being used

    jburto1@hotmail.com (Jeff Burton) wrote in message news:<8a252978.0402061027.651afa91@posting.google.com>...
    > We have a java application on SunOS 5.8 that is attempting access an
    > automounted (amd) NFS resource. I am guessing amd is taking too long


    Why are you using amd and not the native Solaris 8 automounter?

    > to mount the resource, so the application gives up. Since this is
    > SunOS 5.8 the default mount option is NFSv3/TCP, and ?nfsstat ?m' is
    > telling me that all NFS mounts are using v3/TCP . ?nfsstat' is also
    > telling me that there were 0 "connection oriented" timeouts, but that
    > there were a few "connectionless" timeouts. Confusingly, ?nfsstat'


    Is this nfsstat run on the client or the server?

    > says that just over 50% of total nfs calls are v2, all of which are
    > lookup, readdir and readlink. How can over 50% of NFS calls be v2,
    > when everything that is mounted using v3?


    Are you sure nfsstat isn't counting calls to the NFS server that might
    be running on the client?

    It is possible that the automounter often mounts vers=2 (by accident
    perhaps ... say a slow NFS server doesn't respond to the vers=3 ping, but
    then responds to the vers=2 ping).

    Try using the -z option to reset the counters to zero. This will
    verify if vers=2 traffic is being counted against an vers=3 mount.

    You might also explicitly list vers=3,proto=tcp in your
    automounter map entries.

    >
    > I looked at a few other machines in our environment and it looks like
    > all of them are doing the same thing (all v3 mounts, but just over 50%
    > of calls are v2). I looked at the NFS server and it has 1020 UDP
    > calls from the server in question but not one NFS connectionless
    > operation. Why are there so many UDP calls that appear to just be
    > asking for a directory listing, and how/should I stop it from doing
    > that?
    >
    > Thanks,
    > Jeff


  3. Re: ver3 not being used

    spamisevi1@yahoo.com (Mike Eisler) wrote in message news:<36f0f19f.0402121509.261a922b@posting.google.com>...
    > jburto1@hotmail.com (Jeff Burton) wrote in message news:<8a252978.0402061027.651afa91@posting.google.com>...
    > > We have a java application on SunOS 5.8 that is attempting access an
    > > automounted (amd) NFS resource. I am guessing amd is taking too long

    >
    > Why are you using amd and not the native Solaris 8 automounter?


    I should start by saying my mail client blocked your response (which I
    have since fixed). So thinking nobody responded to this, I posted a
    similar message on am-utils@am-utils.org. But for those not on that
    list I will answer these questions here.

    We started running amd many years ago and really need to review wether
    we want to use that or the O/S's native automounter.

    >
    > > to mount the resource, so the application gives up. Since this is
    > > SunOS 5.8 the default mount option is NFSv3/TCP, and ?nfsstat ?m' is
    > > telling me that all NFS mounts are using v3/TCP . ?nfsstat' is also
    > > telling me that there were 0 "connection oriented" timeouts, but that
    > > there were a few "connectionless" timeouts. Confusingly, ?nfsstat'

    >
    > Is this nfsstat run on the client or the server?
    >


    I ran 'nfsstat -cn' on the client. The server (a NetApp) also shows
    the same v2/UDP server connection information.

    > > says that just over 50% of total nfs calls are v2, all of which are
    > > lookup, readdir and readlink. How can over 50% of NFS calls be v2,
    > > when everything that is mounted using v3?

    >
    > Are you sure nfsstat isn't counting calls to the NFS server that might
    > be running on the client?
    >


    Since I ran 'nfsstat -cn' it is, of course, it is not reporting server
    stats.

    > It is possible that the automounter often mounts vers=2 (by accident
    > perhaps ... say a slow NFS server doesn't respond to the vers=3 ping, but
    > then responds to the vers=2 ping).


    That would make sense, but 'nfsstat -m' is showing that the mount is
    v3/tcp. Also when I staticly mount the resource v3/tcp is used
    exclusively.

    >
    > Try using the -z option to reset the counters to zero. This will
    > verify if vers=2 traffic is being counted against an vers=3 mount.


    Did this, and it is that mount point.

    >
    > You might also explicitly list vers=3,proto=tcp in your
    > automounter map entries.


    Did this, didn't help.

    Any other thoughts? Thanks for your response,
    Jeff

    >
    > >
    > > I looked at a few other machines in our environment and it looks like
    > > all of them are doing the same thing (all v3 mounts, but just over 50%
    > > of calls are v2). I looked at the NFS server and it has 1020 UDP
    > > calls from the server in question but not one NFS connectionless
    > > operation. Why are there so many UDP calls that appear to just be
    > > asking for a directory listing, and how/should I stop it from doing
    > > that?
    > >
    > > Thanks,
    > > Jeff


  4. Re: ver3 not being used

    jburto1@hotmail.com (Jeff Burton) wrote in message news:<8a252978.0402171205.696d62a0@posting.google.com>...
    > We started running amd many years ago and really need to review wether
    > we want to use that or the O/S's native automounter.


    I'd try an experiment comparing the native and amd automounters,
    each trial created a 1gbyte file (use mkfile). If you see
    the native automounter doing what is expected, then you need
    serious amd expertise (can't help you there).

+ Reply to Thread