|Date and Author(s)|
All Resource Management Working Group meeting notes should be appended to this document prefaced by a date.
Participants were Brett Bode, David Jackson and Scott Jackson.
It was determined that we will initially have weekly one-hour conference calls on Tuesdays at 2:00 PM CST (12AM PST).
Scott will arrange to have a call in number setup with passcode for the next meeting.
To be connected to the weekly teleconference call 800-664-0771.
Brett filled us in with some of the discussion of the Process Management meeting which he was part of. He made the suggestion that we exert an effort to prepare a focused agenda well before the meeting so that discussions can be kept on track.
It was decided that we would have a rotating meeting chair and rotating meeting secretary -- rotating on 1 month boundaries, with the chair preparing the agenda and leading the meeting and the secretary having the responsibility to take notes and update documents, notebooks, etc. with an overview of the discussion and results.
The secretary during one month will be the chair the following month. Initially we will be starting as follows:
Meeting Chair: Meeting Secretary ------------------- -------------------------- Scott Brett Brett David David ... ...We discussed how we might interact with other working groups and how we will manage the human resources we will each have for the project. We then considered first steps for the project.
We discussed the composition of the resource management pieces and how they fit together. We then discussed the queue manager component and whether the queue manager piece for the meta-scheduler should be the same piece of code as the queue manager for PBS, or if they ought to remain separate.
We each accepted a work item to fill in some of the data in the notebook for our respective areas for the:
Objects and Attributes
Components and Services
We have already had some extra-meeting discussions of XML request/response syntaxes with members of various groups.
Brett Bode, Scott Jackson, and John-Paul Navarro attended.
Scott: Had a meeting with Rusty and Narayan had meeting and need coordination with the Process Management W.G. Suggest being on the Process management WG mailing list.
Scott has been in contact with a group called the New Product Initiative made up of companies including Compaq, HP and Platform computing among others. They are working on standardizing distributed resource management (the emphasize is on distributed). The have first draft documents at:
[I talked with Fred Johnson about involvement with the NPI initiative by this ISIC. See what Fred says on page 9 of main notebook -Al]
They seem eager to have us join them, but membership is $500 or more. There appears to be some overlap in the goals, but also significant differences. Agreed to review the documents on the NPI site and decide via email if we should include them in a conference call next week.
Started discussing how the component breakdown should be done between Scheduler/Queue manage/job manager/process manager. General agreement that there needs to be a "server" component (job manager) that executes the commands such as "Run Job", "Cancel Job", etc, ie the Job manager server component provides command interface between the nodes and the other server software.
Discussed the need for a collaboration tool for use during conference calls. Agreed to give Microsoft Netmeeting a try.
Need another meeting on prototype development.
Expanded components and services on page 6 of the notebook. Defined requirements and responses for each component.
Al brought us up to speed on the PM groups progress on XML schemas.
Al discussed what the Information manager's job is: A) directly provide information, b) pass the request on to the appropriate component, c) respond that the request is invalid. Al will add more detail to the main notebook. One use will likely just be service discovery. This should make the whole system flexible, reducing the need for hardwired locations for each component.
Scott: determine topics of discussion with the other groups.
All: need to flesh out the notebook.
Go over open issues
Notebook table of contents needed. Things are mixed up. We don't have a consistent naming convention.
Maybe produce a document with all of the URL's that we want, so that it's organized how we want it.
What do we think about having a person to take care of organizing stuff.
Everybody should take care of their own files. Nobody wanted to volunteer to keep track of all of the stuff and make sure that the documents were up
Scott to ask Al about putting in the table of contents at the top
Scott J. to come up with an organization of the contents of the web page.
No meeting next week due to SC. We'll wait for the week after.
Go through the action items.
Did people update the dictionary with relevant terms?
Nobody has posted the documentation yet.
Should we have a common document for all of this stuff?
Greg didn't seem to be sure.
What will the dictionaries be used for? (To Eric)
To get the common semantics for use in building the interfaces.
Is it going to be easier to have one document rather than one large document.
What needs to be in there?
Some key words and what they mean.
In two weeks all of the dictionaries should be done. And everyone should take a look and verify if they are done.
What about using some kind of revision control stuff. The electronic notebook is somewhat difficult to use. Things seem to be difficult to change very
well. It’d be better to use something with web access to a directory of files of some kind.
Scott will look at WebDAV or something like that.
XML progress…. Nothing to discuss.
Report on task management queue interface. Rusty was going to give us some status. The major issue was on how the queue and task manager interface etc. When we’ve spoken, issues weren’t very concrete. It’s hard to work on undefined interfaces. When asked about how he wanted to interface, he didn’t seem to care. We were going to create a set of requirements that the task manager needs to support (queue and task manager communication). These were then going to be given to him. Dynamic job issue needs to be taken care of as part of this. Did Brett want to look at that? Brett and David will work on this. Brett will brainstorm and meet with Dave at SC about this.
Integration team. Who’s sitting in on their meetings. JP is talking about the directory service. Kind of like a DNS to have things find each other. It should be very lightweight. Only supposed to provide service connection information.
Information service API. Libraries to access data in the information service. Can the data service serve data to other tools? That problem or idea hasn’t been solved. Can tools query directly to this service of will they be required to go through the service that created the data. To find Queue information do you have to ask the queue manager or can you get that data directly from the information service.
Is there a target timeframe. We’ve discussed the directory service and the information service, but we don’t have a specific timeframe.
Will you ask the lead of the team for some kind of schedule for a prototype. Or even before that if they could define an API so that we could build stuff to interface with that. Depending on the algorithm that we’re using on our code.
Does the app get information from the information service or does it get the information directly.
Does the information service group need information from us? If they’ll let us know, what they need it and when they need it have them let us know.
Packaging and plans:
Is there a plan for how things are going to be packaged? Nobody seemed to know who had this. David Jackson was going to find out if that’s been assigned.
Generation of 1 yr deliverable list.
We need to get a list together to see what needs to be done when so that we can plan on other things that we’re going to depend on and what other people are going to depend on.
How to make a chart of dependencies? How is it workable?
Probably the first cut will need to be a text document.
Dave will try to construct a simple web page with information on what people are responsible for. What do we have that we need to deliver. Everybody should send in a list of what they see to be their tasks and post them to the web and Dave Jackson will collect them into a single document.
Scott Jackson wanted input on job flow stuff. Both local and meta jobs stuff.
Come up with a end-to-end job flow for face to face depicting all of the interactions between components for a typical job (batch, interactive, meta)