OGSA F2F Meeting 17
===================

     Wednesday afternoon, 14 March 2007
     Oracle

* Data grid usecases (Jay, DaveB, 1:30-3pm)

** Participants

  Mark Morgan 
  Jay Unger
  Steve McGough
  Hiro Kishimoto
  Fred Maciel
  Alan Luniewski
  Ellen Stokes
  Andrew Grimshaw
  Chris Kantarjiev
  Michel Drescher 
  Andreas Savva

  Dial in: Chris Smith, Dave Berry

  Minutes: Andreas Savva

  - Long standing action; not much progress
  - Jay has talked with some projects (condor etc)
  - Platform has some support but not pre-configured; for most users
    making sure that data is pre-placed is good enough
  - Univa has published something related (data movement
    orchestrator); eventually might be on-the-fly, dynamically
    orchestrated replication. 
    - Done for commercial entities; not just for escience applications
  - Current practice is not very advanced
  - Data as first class resource
  - Better to go back to basics; how people like to think about these
    entities, what I/O paradigms have served community and are well
    accepted.
    - Replication might be under the covers and transparent
    - Access method for applications: what they already use
      - Method varies depending on application; need to have those
        different approaches
  - How is advertisement done and what criteria are used to narrow
    down data sets published?
    - How many entities do you publish? All or some subset, and how
      that subset is determined? Concern with the size of this space,
      and RNS applicability or not.
    - How is the search space (for a evaluating xquery requirement)
      determined
    - Is this applicable for data? If there is a reference (e.g., EPR)
      to the needed data is it enough?
  - Recapitulated that state of the art is to pre-place data and
    assume it is there; only a small number of users actually do
    anything more refined like advertising.
  - Job submission as an example: how should 'files' or 'data' be
    modeled. Is it just a requirement on a resource or something
    different (more data specific). What information is needed by the
    scheduler to make sure that something is available; and whether
    there is an objective to optimize.
  - Describe requirements, e.g., what files are required. How the
    requirements are satisfied depends on the state of the system,
    e.g., what might already be available locally and the capabilities
    of the 'container'
    - There are a number of requirements on other areas and those
      requirements would also have to be documented.

  This action is closed.

  AI-0314pm-a: Jay (and Andrew) to write up a short paper describing the
               resolution of this action

* QoS discussion (Steve, Andrew, 3:15-5pm) 

** Participants

  Mark Morgan 
  Jay Unger
  Steve McGough
  Hiro Kishimoto
  Fred Maciel
  Alan Luniewski
  Ellen Stokes
  Andrew Grimshaw
  Chris Kantarjiev
  Michel Drescher 
  Andreas Savva
  Dave Snelling
  Donal Fellows

  Dial in: Fred Brisard

  Minutes: Andreas Savva

  - Andrew's presentation; no definition, just trying to identify what
    it is that it means. Seems closer to what SLA is used for, rather
    than QoX
  - Continuous vs discrete approach; providing a set of discrete
    options
  - Need to factor in cost otherwise everyone would ask for maximum
    values of all dimensions
  - WS-Agreement shortcomings (just a protocol; need terms) and
    problem of not having negotiation
  - Need for terms to describe SLAs, as a start. Not necessarily
    within OGSA-WG. A workshop style might be one idea.
  - Andrew volunteered to arrange a session at OGF20 to discuss this
    topic. NextGRID is also doing some work in this area.

  AI-0314pm-b: Andrew will contact stakeholders with a view to
               arranging a session at OGF20, probably as an ad-hoc
               BOF. Choose a specific topic in the SLA space (terms on
               response time, cost, availability, dependability) that
               can be applied to BES and ByteIO.

  This could later lead to a workshop to define a broader set of terms.