Conference humane::scheduler

Try a $ schedule check/all and look again to the output of sched sh stat.
What does schedule sh load tell you?

eric

    Hi Eric,

    Thanks for your help.  I will onsite again tomorrow.  From the findings
    last time, I notice that when showing the logicals NSCHED*, the
    logical NSCHED$LBAL$CPU_WEIGHT is not shown by default, even the
    NSCHED$DEFAULT_JOB_MAX, NSCHED$DEFAULT_JOB_PRI or NSCH$LBAL$INTERVAL
    are not shown.  Is it normal?  If it is not shown, does it mean the
    values are use the default one?

    Thanks again for your help.

    Kenneth Leung

    nsched$lbal$cpu_weight does not need to be defined unless you want to
    fine-tune the algorithm by which a scheduler calculates the service
    ratings used by load balancing.   There is a memory caluclation and a
    cpu calculation that are plugged into a formula. If set, 
    nsched$lbal$cpu_weight lets you affect the multiplier used for the cpu
    portion of the rating on that node.   One use for this is if you have
    cluster nodes that are architecturally different from each other (for
    example a 2 node cluster consiting of a node with 6-6 VUP CPU's and
    another with 1-36 VUP CPU.   The cpu portion of the service ratings
    for the 6-6Vup node will be almost as high if there are more job slots
    available, and you may not want that, so you can tune it (without
    having to go to custom load balancing, which has much more overhead
    since it involves spawning a process before each job is run, to execute
    the script that defines which node it should run on).  
    
    
    All the load balancing logicals have defaults if they are not set 

    Does the command 
    
    	$ sched set load on
    
    fix it ?

Hi,

I have set the logical NSCHED$LOAD_BALANCE to "on" in the SCHEDULER$STARTUP.COM,
does it equivalent to "$ sched set load on"?

I find that if I "$ sched set load off", there is no "Rating" displays.
I see the rating displays when I show status.  So I think we have already
turn on load balancing, is that right?

By the way, refer to ".1", during my last onsite, I issue the command 
"$ schedule check/all", the jobs still submit to the default node.  

Thanks for your help again.

Kenneth Leung

    The nsched$load_balance logical is looked by the first scheduler in the
    cluster started, to see whether it should come up with load balancing
    ON or OFF.  Other schedulers that come up later copy their setting from
    the "default" scheduler,
    whereas the $ sched set load {on off} command is dynamic - it sends a
    broadcast message to all schedulers telling them to switch their
    setting to ON or OFF.
    
    If you see ratings numbers in the user interface, load balancing
    "should" be on, otherwise something is in an inconsistent state.
     
    You should try the $sched set load on  command anyway, it will do no
    harm if load balancing is already on.   The best way to really find out
    what is happening is to stop all the schedulers in the cluster, start
    one scheduler on a hard copy terminal or window where output is logged,
    (it will become the "default" scheduler since it is started first - the
    "default" assigns jobs to nodes for load balancing), then start another
    scheduler on a node with a higher rating.   Then watch the debug output
    to see what's happening.   Alternatively, if you don't want to mess
    with this, you can do    $ SCHED SET DEBUG ON/NODE=default_node
    run a few jobs that should be load balanced to another node, then
    $ SCHED  SET DEBUG OFF /node=           Look at the scheduler's output,
    probably nsched$:nsched.log unless you've moved it.    (Don't leave
    debug mode on permanently, the log file will get very big and hurt
    performance, you should be able to see very quickly what's happening by
    looking at the end of the file)