|Date and Author(s)|
Meeting minutes for 12/18/01
Scott Jackson PNNL
David Jackson PNNL
Brett Bode Ames Lab
Eric Debenedictus SNL
Stephen Scott ORNL
Al Geist ORNL
In opening minutes of call, Scott mentioned progress on creation of prototype for Allocation manager which handles queries and responses via XML using the half closed socket according to the straw vote in the last face-to-face meeting. Scott created a simple draft schema and will work in the next couple of weeks to get it to validate.
Eric brought up issues with the use of half-closed sockets at least for the long term since it did not support multiple requests per socket among other things. I think we concluded that this was only meant to be a super-simple implementation to get things going and it is certain to change later after we get over the communication and parsing hurdle, and that component creators are free to code other implementations as long as they also support this one.
Discussed vision statement: Will change terascale to tens of thousands of nodes
Proposed that XML namespace be of the form:
http://www.scidac.org/ScalableSystems/component i.e. http://www.scidac.org/ScalableSystems/pm /* Process Manager */ http://www.scidac.org/ScalableSystems/qm /* Queue Manager */ http://www.scidac.org/ScalableSystems/am /* Allocation Manager */ http://www.scidac.org/ScalableSystems/sched /* Scheduler */
It was agreed that Scott will take over the responsibilities for creating and distributing the weekly meeting agenda. Decided that we will rotate working group secretary in 3 month (quarterly) terms to take and post meeting minutes. Gary Skouson (PNNL) had previously agreed to be secretary. I will see if he will continue through the remainder of this quarter (quarter delimited by quarterly face to face meetings).
Brett briefly discussed his notebook document entitled:
RMS queries to PMWG components
And requested that it be reviewed and comments returned.
It proposes two communication mechanisms, a query/response mechanism and a subscription/notification mechanism.
Scott brought up a suggestion that perhaps a streaming connection where the socket remains open is a third mechanism Ð while some did not recognize the distinction between subscription and streaming, particularly if the clients keep a socket open continuously with the server.
Brett also published another document to the notebook: Common RMS attributes
We talked at some length about the XML schema fragments inside the document seeking clarifications and offering insight.
We discussed the strategy of starting with a more loosely defined schema in the initial phases while rapid prototyping is carried out, with schema validation increasing as the interfaces become more tested and concrete.
Began discussion of getting prototypes to communicate with each other: AM-Sched, QM-Sched, QM-PM.
We will have modified meeting schedules through the holidays:
4EST Wed December 26th
3EST Wed January 2nd