This is a static archive of the previous Open Grid Forum GridForge content management system saved from host forge.ogf.org file /sf/wiki/do/viewPage/projects.gsa-rg/wiki/SchedulerInteractions at Fri, 04 Nov 2022 22:34:14 GMT SourceForge : View Wiki Page: SchedulerInteractions

Project Home

Tracker

Documents

Tasks

Source Code

Discussions

File Releases

Wiki

Project Admin
Search Wiki Pages Project: GSA-RG     Wiki > SchedulerInteractions > View Wiki Page
wiki1617: SchedulerInteractions

Scheduler Interactions


Synopsis

It has been decided to start with a simple use case where two schedulers interoperate with each other. The goal is to show that it is possible that one scheduler (the primary scheduler), which decides (how is out of scope) that it cannot fulfil a scheduling request, passes this request to another scheduler (the secondary scheduler) for processing. The second scheduler then checks its capabilities and, in case it can fulfil the request, tells the first scheduler under which conditions this is possible. Both then agree (or disagree) on the conditions to fulfil the request. Potentially it is possible to re-define these conditions before an agreement is reached. At the end of the process, the first scheduler knows that the second one will fulfil the request and it knows under which conditions this will be done.

Within scope

  • Simple request for a computing resource
  • Basic set of service-level objectives (SLOs) passed from one scheduler to another (mainly time-related)
  • Definition of interfaces
  • Defintion of messages

Out of scope (during the first period)

  • Other resource types (e.g. integration of data resources)
  • Complex service-level objectives
  • Definition of a unified information model (in general it is assumed that the different local Grid Information Systems have different information models)

Architecture

Main entities

  • Client
  • Scheduler
  • Adapter

Utility entities

  • Registry
  • Local GIS (Grid Information Service)

inter_scheduler_services.jpg


Approach

The client sends an JSDL request to the primary scheduler. The primary scheduler decides (how is not of interest) that it cannot fulfil the request, and starts negotiating with the secondary scheduler whether it can fulfil the request and if so, under which conditions. This is done using an SLA, in our case the one specified by WS-Agreement. Since we want to be able to not only execute a two-phase commit, the current protocol specified by WS-Agreement will not be enough and we therefore discuss an extension.

Assumptions

  • A scheduler does not know anything about other schedulers. All information needed has to be passed via the one scheduler interoperation interface.
  • There is no common information model (apart form CIM, but this is not meant here).

JSDL "profile"

tbd

Scheduling attributes

  • Hard deadline for the end time of job execution
  • Earliest start time
  • Duration
  • Service-level objectives/scheduling preferences/scheduling objectives: xsd:string or xsd:any.

Additional attributes for inter-scheduler communication

Negotiation protocol

tbd (until now see discussion at issue_008).

Actions

  • (Ariel) Starting point for the JSDL profile and the scheduling attributes. The goal is to agree on a basic set to be put into an SLA.
  • (Oliver) Write a document on the negotiation protocol.

Issues

Open

  • issue_001 Definition of a real use-case: which schedulers (VIOLA, GRMS, others?), first ideas concerning scenario, what should be the exact outcome we want to demonstrate?
  • issue_002 Design of scheduler interactions (sequence diagrams?): probably we can distinguish several scenarios depending on assumed environment, type of Grid etc.?
  • issue_003 Between which entities should agreements be establish?
  • issue_004 What are the parameters needed for scheduling? Parameter semantics/syntax?
  • issue_005 How to handle different models of constraints/preferences?
  • issue_006 Job execution: Who is executing a job and who is controlling/managing/monitoring it?
  • issue_007 Does the primary scheduler/broker transfers the responsibility for the whole job to the secondary one or is a splitting possible?
  • issue_008 Negotiation protocol.

Closed

  • During the 2006/10/04 telecon it has been decided that it would be a good idea to use the generic communication stages (see slide 2 of this PPT) and start with sequence diagrams for the different sub-steps needed to perform scheduling.
  • The decision whether or not a scheduler passes a request to another scheduler in case it cannot fulfil it is internally taken and not of concern wrt to the scenario (2006/10/23).
  • Data management very much of interest, but will not be tackled in the first period (2006/10/23).

Attachments:
inter_scheduler_services.jpg [SchedulerInteractions/inter_scheduler_services.jpg]
 



Versions Associations Attachments (1) Back Links  
Version Version Comment Created By
Version 20 Philipp Wieder - 10/23/2006
Version 19 Philipp Wieder - 10/23/2006
Version 18 Philipp Wieder - 10/23/2006
Version 17 Philipp Wieder - 10/23/2006
Version 16 Philipp Wieder - 10/23/2006
Version 15 Philipp Wieder - 10/23/2006
Version 14 Philipp Wieder - 10/23/2006
Version 13 Philipp Wieder - 10/23/2006
Version 12 Philipp Wieder - 10/23/2006
Version 11 Philipp Wieder - 10/23/2006
Version 10 Philipp Wieder - 10/23/2006
Version 9 Philipp Wieder - 10/23/2006
Version 8 Philipp Wieder - 10/23/2006
Version 7 Philipp Wieder - 10/23/2006
Version 6 Philipp Wieder - 10/12/2006
Version 5 Philipp Wieder - 10/04/2006
Version 4 Philipp Wieder - 10/04/2006
Version 3 Philipp Wieder - 10/04/2006
Version 2 Philipp Wieder - 10/04/2006
Version 1 Philipp Wieder - 10/04/2006



The Open Grid Forum Contact Webmaster | Report a problem | GridForge Help
This is a static archive of the previous Open Grid Forum GridForge content management system saved from host forge.ogf.org file /sf/wiki/do/viewPage/projects.gsa-rg/wiki/SchedulerInteractions at Fri, 04 Nov 2022 22:34:31 GMT