Reply retransmits in TCP - NFS

This is a discussion on Reply retransmits in TCP - NFS ; Hello, I wrote a little NFS server as a proxy-server... The problem is, when registering the NFS server on a TCP transport, it comes to retransmits of reply (!) messages and the server gets very slow. In UDP, everything works ...

+ Reply to Thread
Results 1 to 8 of 8

Thread: Reply retransmits in TCP

  1. Reply retransmits in TCP

    Hello,

    I wrote a little NFS server as a proxy-server...

    The problem is, when registering the NFS server on a TCP transport, it
    comes to retransmits of reply (!) messages and the server gets very slow.
    In UDP, everything works correctly...

    Anybody has a hint for me?

  2. Re: Reply retransmits in TCP

    Jochen Witte wrote in message news:...
    > Hello,
    >
    > I wrote a little NFS server as a proxy-server...
    >
    > The problem is, when registering the NFS server on a TCP transport, it
    > comes to retransmits of reply (!) messages and the server gets very slow.
    > In UDP, everything works correctly...
    >
    > Anybody has a hint for me?


    You don't provide much to go on.

    Why is your server retransmitting responses? They should be
    sent once. Of course, at teh TCP level they could
    be re-sent if the client's TCP end point
    is not responding. What does a packet trace show you?

  3. Re: Reply retransmits in TCP

    In the trace just the duplicate reply shows up. On the TCP level however,
    I see an ACK from the server before and an ACK from the client immediatly
    after the retransmitted reply (reproducible).


    Am Sun, 20 Jun 2004 06:44:22 -0700 schrieb Mike Eisler:

    > Jochen Witte wrote in message news:...
    >> Hello,
    >>
    >> I wrote a little NFS server as a proxy-server...
    >>
    >> The problem is, when registering the NFS server on a TCP transport, it
    >> comes to retransmits of reply (!) messages and the server gets very slow.
    >> In UDP, everything works correctly...
    >>
    >> Anybody has a hint for me?

    >
    > You don't provide much to go on.
    >
    > Why is your server retransmitting responses? They should be
    > sent once. Of course, at teh TCP level they could
    > be re-sent if the client's TCP end point
    > is not responding. What does a packet trace show you?



  4. Re: Reply retransmits in TCP

    Jochen Witte wrote in message news:...
    > In the trace just the duplicate reply shows up. On the TCP level however,


    if all you see ia a duplicate and not the riginal, then how do you know
    it is a duplicate?

    :-)

    > I see an ACK from the server before and an ACK from the client immediatly
    > after the retransmitted reply (reproducible).


    if your server is really sending multiple replies and the client isn't
    sending multiple requests, then there's a bug in either your
    RPC library or your server (are you calling svc_sendreply()
    multiple times?)

    >
    >
    > Am Sun, 20 Jun 2004 06:44:22 -0700 schrieb Mike Eisler:
    >
    > > Jochen Witte wrote in message news:...
    > >> Hello,
    > >>
    > >> I wrote a little NFS server as a proxy-server...
    > >>
    > >> The problem is, when registering the NFS server on a TCP transport, it
    > >> comes to retransmits of reply (!) messages and the server gets very slow.
    > >> In UDP, everything works correctly...
    > >>
    > >> Anybody has a hint for me?

    > >
    > > You don't provide much to go on.
    > >
    > > Why is your server retransmitting responses? They should be
    > > sent once. Of course, at teh TCP level they could
    > > be re-sent if the client's TCP end point
    > > is not responding. What does a packet trace show you?


  5. Re: Reply retransmits in TCP


    Am Tue, 06 Jul 2004 17:18:37 -0700 schrieb Mike Eisler:
    > Jochen Witte wrote in message news:...
    >> In the trace just the duplicate reply shows up. On the TCP level however,

    >
    > if all you see ia a duplicate and not the riginal, then how do you know
    > it is a duplicate?
    >
    > :-)
    >
    >> I see an ACK from the server before and an ACK from the client immediatly
    >> after the retransmitted reply (reproducible).

    >
    > if your server is really sending multiple replies and the client isn't
    > sending multiple requests, then there's a bug in either your
    > RPC library or your server (are you calling svc_sendreply()
    > multiple times?)
    >
    >>
    >>
    >> Am Sun, 20 Jun 2004 06:44:22 -0700 schrieb Mike Eisler:
    >>
    >> > Jochen Witte wrote in message news:...
    >> >> Hello,
    >> >>
    >> >> I wrote a little NFS server as a proxy-server...
    >> >>
    >> >> The problem is, when registering the NFS server on a TCP transport, it
    >> >> comes to retransmits of reply (!) messages and the server gets very slow.
    >> >> In UDP, everything works correctly...
    >> >>
    >> >> Anybody has a hint for me?
    >> >
    >> > You don't provide much to go on.
    >> >
    >> > Why is your server retransmitting responses? They should be
    >> > sent once. Of course, at teh TCP level they could
    >> > be re-sent if the client's TCP end point
    >> > is not responding. What does a packet trace show you?


    Well: I can identify duplicate replies out of the XID, which is the same
    for all replies
    JOchen



  6. Re: Reply retransmits in TCP

    Jochen Witte wrote in message news:...
    > Am Tue, 06 Jul 2004 17:18:37 -0700 schrieb Mike Eisler:
    > >> > Why is your server retransmitting responses? They should be
    > >> > sent once. Of course, at teh TCP level they could
    > >> > be re-sent if the client's TCP end point
    > >> > is not responding. What does a packet trace show you?

    >
    > Well: I can identify duplicate replies out of the XID, which is the same
    > for all replies


    I suspect the TCP sequence number is the same too. :-)

  7. Re: Reply retransmits in TCP

    Mike Eisler wrote:
    > Jochen Witte wrote in message news:...
    >> Well: I can identify duplicate replies out of the XID, which is the
    >> same for all replies


    > I suspect the TCP sequence number is the same too. :-)


    And if it is not, it means the NFS stuff is indeed resending things.
    I guess that would indeed be a bug if it was the reply.

    rick jones
    --
    a wide gulf separates "what if" from "if only"
    these opinions are mine, all mine; HP might not want them anyway...
    feel free to post, OR email to raj in cup.hp.com but NOT BOTH...

  8. Re: Reply retransmits in TCP

    Rick Jones wrote in message news:...
    > Mike Eisler wrote:
    > > Jochen Witte wrote in message news:...
    > >> Well: I can identify duplicate replies out of the XID, which is the
    > >> same for all replies

    >
    > > I suspect the TCP sequence number is the same too. :-)

    >
    > And if it is not, it means the NFS stuff is indeed resending things.
    > I guess that would indeed be a bug if it was the reply.


    Unprecedented in my experience, but yes it would be a bug.

+ Reply to Thread