"too many failed RPC calls" during the SFS benchmark - NFS

This is a discussion on "too many failed RPC calls" during the SFS benchmark - NFS ; Hello, I am running SPEC SFS97_R1 V3.0 benchmark and getting a lot of the following errors: sfsNN: too many failed RPC calls - M good K bad where N,M,K are numbers Does it make the benchmark results invalid? Does it ...

+ Reply to Thread
Results 1 to 3 of 3

Thread: "too many failed RPC calls" during the SFS benchmark

  1. "too many failed RPC calls" during the SFS benchmark

    Hello,
    I am running SPEC SFS97_R1 V3.0 benchmark and getting a lot of the
    following errors:

    sfsNN: too many failed RPC calls - M good K bad
    where N,M,K are numbers

    Does it make the benchmark results invalid?
    Does it point to my NFS server problem? Which problem?
    TIA,
    -- Leon


  2. Re: "too many failed RPC calls" during the SFS benchmark


    "napobo3" wrote in message
    news:1135432849.259919.265940@g47g2000cwa.googlegr oups.com...
    > Hello,
    > I am running SPEC SFS97_R1 V3.0 benchmark and getting a lot of the
    > following errors:
    >
    > sfsNN: too many failed RPC calls - M good K bad
    > where N,M,K are numbers
    >
    > Does it make the benchmark results invalid?
    > Does it point to my NFS server problem? Which problem?
    > TIA,
    > -- Leon
    >


    Leon,

    1. Does it make the result invalid.... YES...Invalid.
    2. Does it point to my NFS server ... Possibly..
    Server failing RPCs, or network is munging them,
    or NIC is munging them, or memory corruption, or
    CPU cache corruption... Something is causing the RPCs
    to fail.
    3. Why would a license holder of SPEC's SFS benchmark be
    going to the news groups, instead of just asking SPEC
    for support ? Seems strange to go fishing for an answer
    in the wild, when there is a formal support path ??
    Why would one do this ?

    BTW: The SFS benchmark lets you get away with a few
    bad RPCs, it only complains when the number of bad
    RPC's is outrageous....

    Enjoy,
    Postmaster



  3. Re: "too many failed RPC calls" during the SFS benchmark

    Postmaster wrote:
    > "napobo3" wrote in message
    > news:1135432849.259919.265940@g47g2000cwa.googlegr oups.com...
    >
    >>Hello,
    >>I am running SPEC SFS97_R1 V3.0 benchmark and getting a lot of the
    >>following errors:
    >>
    >>sfsNN: too many failed RPC calls - M good K bad
    >>where N,M,K are numbers
    >>
    >>Does it make the benchmark results invalid?
    >>Does it point to my NFS server problem? Which problem?
    >>TIA,
    >>-- Leon
    >>

    >
    >
    > Leon,
    >
    > 1. Does it make the result invalid.... YES...Invalid.


    If you look at the sfssum file, it will say if the result
    is within tolerence. Some failures are acceptable, but not
    too many.

    > 2. Does it point to my NFS server ... Possibly..
    > Server failing RPCs, or network is munging them,
    > or NIC is munging them, or memory corruption, or
    > CPU cache corruption... Something is causing the RPCs
    > to fail.


    SFS does not check the contents of the reply when testing
    performance, only during validation does it validate
    the replies. An RPC failure is simply when a request was sent
    without receiving a reply within a small amount of time.
    Why would this happen? Under UDP it could be the client,
    network, or server dropping the request or the reply. Under
    TCP it is usually the server is too overloaded and the
    reply is too late.

    > 3. Why would a license holder of SPEC's SFS benchmark be
    > going to the news groups, instead of just asking SPEC
    > for support ? Seems strange to go fishing for an answer
    > in the wild, when there is a formal support path ??
    > Why would one do this ?


    How about support@spec.org they pay people to answer these
    types of questions which is expected since you paid them
    for the benchmark.

    -David

+ Reply to Thread