Minutes of the GGF 17 GSA-RG session
------------------------------------

[Please see slides on GridForge for details]

- Agenda bashing

- Roadmap presented

- Status update

- Interaction between Scheduling Instances
 * Motivation: Not only one central Grid scheduler -> communication between different 

systems needed
 * Use case presented -> Discussion should result in requirements for an interaction 

interface and protocol
 * Discussion & collection of initial questions/issues:
  + [Ramin] Some people think that contacting an information service and getting 

information about load etc. is enough to schedule. But interesting information is often not 

exposed by the service provider (like e.g. possible start time etc.).
  + [Ariel] The kind of information needed will be use case-dependent. Should this be 

reflected in the interface/protocol or should we try to find an abstract solution?
  + [Ariel, Question] Is there just one comm. channel or does a scheduler of VO A is also 

contacting an information system at VO B? - [Ramin, Wolfgang] The latter scheme fits our 

projects. 
  + [Wolfgang] Description of VIOLA scheduler to adapter/scheduler comm.
  + [Philipp] Knowledge base needed to communicate between Grid Scheduling instances? How 

do we capture semantics?
  + [Ramin] Is it possible to have a GSA scheduler resource type agnostic?
  + Technical details for a prototype to drive requirements on inter-scheduler 

requirements. Follow this in parallel to the higher-level discussion (document two and 

three).
   # Basic job: JSDL plus extensions
   # Negotiation: WS-Agreement plus Negotiation protocol
   # Something else?
  + OGSA-RSS: Not a full coverage of the whole job lifecycle like negotiation
  + Existing projects/collab.s with interest in scheduler interop.:
   # Two schedulers in D-GRID (Uni Dortmund)
   # GMS (PSNC)
   # VIOLA (Fraunhofer SCAI, FZJ; may be basis for a UNICORE solution)
   # Calana (Fraunhofer ITWM; second half of 2006)
  + [Ramin] There will be different scheduling policies which have to be considered.
  + Semantics
   # [Philipp] It is essential to discuss what kind of information is transferred via the 

inter-scheduler protocol, which have to be interpreted/understood by the different 

schedulers.
  + Are there commonalities between data-oriented and compute-oriented resource 

management/scheduling? Are enough people interested in that? [Currently Uni Dortmund, 

Fraunhofer SCAI and ITWM end of the year]
  + [Question] Is monitoring within the scope? - [Answer, Ramin] Start with the part until 

the actual execution of the job. Monitoring will be added later.
 * Next steps:
   + [Question, Ariel] Start from analysis of existing systems or abstract interface? - 

[Answer, Ramin] One slide with an idea, solution, vision from everybody. Concentrate on the 

phase(s) before the actual submission.
   + Review the ideas. -> Depending on the review we can discuss how to proceed.
  
- Presentation by Mathias Dalheimer - Calana Grid system [slides at GridForge]

- Next telecon June 7, 2006, 09:00 UTC