Summary PHYSGI Meeting 4/18/94

                                                         B. Gibbard


    Next meeting will be Monday, April 25, 11:30 - 12:00, in Room 2-160.


    An effort will be made to get the PHYSGI01 WWW home page pointed to by
    other WWW servers in the BNL community soon.  In the mean time one can
    access it by opening:  http://www.phy.bnl.gov.


    Major users for the week beginning 4/18/94 are expected to be:

        1)  E871 (Worm et al.) - 3 CPU intensive Monte Carlo jobs

        2)  E881 (Hackenberg et al.) - 6 CPU intensive data analysis jobs 

        3)  E891 (Saulys et al.) - 1 CPU intensive Monte Carlo job.

        4)  E850 (Gushue et al.) - 4 CPU intensive analysis jobs by Zack
                                   2 perhaps not very CPU intensive jobs
                                                 for EVA calibration

        If the above jobs are all simply allowed to compete 1) the machine
        will swap hard, wasting cycles, 2) interactive users will get terrible
        keyboard response, 3) no job will likely get more than 1/2 of a CPU
        so all jobs will take at least twice as long as necessary to complete.


    Plan for week beginning 4/11/94:

 ****** There are new rules for running this week ******

        First a definition:  A production job is defined as any job which will
        accrue an hour or more of CPU time.
        
        Each experiment is still allowed to run up to 6 concurrent "production"
        jobs but all except 1 of those six must be run "niced" to "1".  It is
        possible that a decision may be made during the week to change this
        "nicing" level.  If so everyone will be informed.

        This "nicing" scheme goes into effect 8:00 am Tuesday morning,
        April 18, and remains in effect until further notice.  Starting
        Tuesday morning no experiment should be running more than one
        production job which is not "niced".

        It is hoped that this will lead to a running configuration in which
        each experiment is able to run one process so that it gets a major 
        fraction of a CPU and thus completes on a reasonable time scale and
        that interactive users will get reasonable response.  Additional CPU
        will then be shared between the "niced" jobs.


    Status of problems from the Upgrade to IRIX 5.2:

        CASEVISION debugger is now working.

        Clock irregularity problem has been worked around.

        EScall is broken under IRIX 5.2 and SGI is on notice.

        FORTRAN has a serious bug (Arrays in Structures can not be properly
        passed with a variable index).  SGI is on notice that this is a
        serious problem.

        Tapeman has been fixable and now works as it did under IRIX 4.05


    During a discussion of the allocation of disk on PHYSGI it was found that
    there was only a single request for additional disk space, that from E850.
    This request can be satisfied so at this point there is no need to
    reevaluate past allocations of disk.  Such a need could appear in the
    future.