[Info-vax] counting-lock ?

[Bob Koehler]

In article <hbqldi$b6g$2 at pcls4.std.com>, moroney at world.std.spaamtrap.com (Michael Moroney) writes:
> Marty Kuhrt <marty at spamloop.kuhrt.net> writes:
> 
>>Set up a special queue for the resource and set the queue job limit 
>>(/JOB_LIMIT=5).  If something is submitted, and five are executing, it 
>>will wait.  If a job completes or dies (assuming it doesn't hang), then 
>>the next job in line gets run.
> 
> Does anyone know what algorithm is used by the queue manager to implement
> this (allow JOB_LIMIT jobs to run but no more, and when one completes,
> select the one waiting the longest to run) ?

   The queue manager is still a single process, one per cluster,
   although it can migrate from node to node.  It's trivial for a single
   process to keep track of these tings itself.  In the past there
   have been several changes to the queue manager to keep reasonable
   performance on large clusters.

   Processes started by the queue manager can, and do, run on many
   nodes.  Which is how batch and print queues are be "on" various nodes.