Subject: More detail from Steve Ellis
Date: Wed, 07 May 1997 14:12:33 EDT
From: "Naomi B. Schmidt" <nschmidt@MIT.EDU>


------- Forwarded Message

Received: from SOUTH-STATION-ANNEX.MIT.EDU by po6.MIT.EDU (5.61/4.7) id AA10017; Fri, 23 Jun 95 22:57:26 EDT
Received: from BOLOGNESE.MIT.EDU by MIT.EDU with SMTP
	id AA25919; Fri, 23 Jun 95 22:57:25 EDT
From: ellis@MIT.EDU
Received: by bolognese.MIT.EDU (5.57/4.7) id AA27964; Fri, 23 Jun 95 22:57:24 -0400
Message-Id: <9506240257.AA27964@bolognese.MIT.EDU>
To: nschmidt@MIT.EDU
Subject: Batch queue background
Date: Fri, 23 Jun 1995 22:57:22 EDT


Naomi

   I won't be able to do the ME batch requirements analysis
until I get back, but the following is an outline of the
capabilities of the Load Balancer software that I am using
to run a batch environment on 3 IBM workstations, 1 SGI
and 1 DEC Alpha (planned).  The system consists of
4 major parts:

   a. a master daemon to coordinate job dispatch to an
      available cpu that meets the job request criteria
      The master daemon maintains a single job queue for all
      requests and then matches requests with resources available.

   b. a local daemon running on each cpu that is providing
      cpu service. This daemon runs jobs and handles I/O, errors
      and enforcing appropriate limits.

   c. a configuration file that defines 3 major areas.
      First, it is used to identify machines available by time of day, 
      load limits, resources such as RAM  and swap, and jobs that can 
      run on each. This file also sets parameters for users and groups. 
      For example, the default environment passed to a user job can be 
      defined here. Finally, the configuration file specifies the 
      structure of the batch queue by assigning priorities and limits
      to job requests. A few examples:

        1. workstations can be declared as available if idle or available
           for set times.

        2. rules for what to do with jobs when the workstation is no longer
           available include kill, stop, renice, requeue.

        3. groups based on courses can be given higher priority, longer cpu
           time limits or access to specific software 
 
      One last function of the configuration file is to provide license 
      function for the software. The hardware running the master daemon must
      be registered with the vendor.      

   d. A set of user and manager utilities to queue up jobs, get status
      of jobs, adjust queued job characteristics and kill jobs.


One of the key features of the configuration file for Athena usage is
that a wrapper program can be specified so that each user job request
is passed to the wrapper program as command line arguments. This program
provides the necessary hook for any Athena required tasks such as 
authentication, locker attachment and environment setup. An initial look at
the setup indicates that this wrapper is the main required development
effort required to use this software in the Athena environment.  


Steve
:wq

zz   



------- End of Forwarded Message