This is a static archive of the previous Open Grid Forum Redmine content management system saved from host redmine.ogf.org file /dmsf_files/7595?download= at Fri, 04 Nov 2022 18:40:58 GMT
Minutes of CGS-WG Meeting at GGF7
Tokyo, Japan
6 March 2003
Carl
Kesselman ran the meeting.
Draft doc has been circulating, describing job
submission interfaces. Does not included detailed modeling of resources or
policies.
Started with Andrea Westerinen's demo, showing the JSIM MOF in
action. Andrea's presentation is
here.
Issue
raised by Andrea about whether Batch Service is always what is responsible for
taking jobs off the queue. Carl says must support brokers that accept jobs, then
send them to specific batch queues. We need to work more on priorities for
job submission.
Ellen listed a few more issues. Heidi Neumann had
previously proposed some PBS issues. Is ExecutionCandidate association in
the right place? Carl: 2 use cases (or more). To do brokering/selection, I look
at a bunch of different services and decide which to use, and look at policy,
priority, etc. If I have a 100-processor job that requires a Gig of memory, no
point to submit to a service that cannot honor that. Also, a service can
maintain multiple queues. Maybe ExecutionCandidate should be raised up and moved
to BatchService. (Moved in real-time on screen by Ellen Stokes!)
Heidi's other question was whether the BatchJob class should provide a
description of the task to be executed. Group decision is that this is
probably opening Pandora'a box and is out of scope.
Carl's use cases:
manageability case, if something fails; discovery case; single submission case;
brokering case. Same model should work for all.
Tom Roney gave a demo
non-specific to JSIM, centered around using CIM to effectively manage
systems. Some details:
Demo:
Control Console
Event Subscription
Automated actions
See: http://www-124.ibm.com/sblim/
Write to: troney@ncsa.edu OR sblim@ncsa.edu
Closing from Carl:
We need to move the document forward. In the interest of time, we should try to schedule a face to face meeting before GGF8. This should hopefully allow us to finalize the document for GGF8.
We should setup a Bugzilla database to track issues. Larry will do this.
We have an end of March deadline to send the JSIM schema to DMTF for the 2.8
release, but we can change later on with implementation
experience.