Conference vmszoo::rms_openvms

3011.0. "Sequential read takes different tim on VAX & Alpha" by COL01::VSEMUSCHIN (Duck and Recover !) Mon Feb 17 1997 12:55

    I have following problem:
    
    Program A writes .LOG file: 
      1. .LOG file is opened only when there is record to write in
      2. .LOG file is opened for shared append (I think, that FORTRAN
         attributes are converted to FAB$V_SHRPUT)
      3. After the record is appenden .LOG file is closed
    Program B types the last .LOG record
      1. .LOG file is opened for shared read
      2. all .LOG file records are readed sequntially
      3. las one is typed
    
    Cluster consist of one VAX node and one Alpha node.
    There are many programs of type A that are running on VAX node (and only
    there !). If program B is started on VAX node it need, let's say, 1 minute
    to poll all .LOG files. On Alpha node it needs 5-6 min.
    
    I wrote test program that creates random number of sequential files and
    then read them in random order. On Alpha it takes little less time as
    on VAX. Then I said customer that it is obviously not file system or
    RMS performance problem, but locking problem. But they insist, that
    this time difference 1:5 is kept even if all A programs are inactive
    and don't write any information in .LOG files ?
    
    Any ideas ?
    
    TIA Seva
    
    PS I will soon wisit the customer site so I can provide more
    information. 
    
    :-$ almost forget. .LOG files are called after process name, so they
    use sys$process_scan to find all active A processes and then open
    all 'active' .LOG files. It is possible that $process_scan works
    slowly, when looking for processes on another node ... But I don't
    expect that it is THE bottleneck.
    
    =Seva