Northwestern University

1
Group 2 report

• Access Optimizations to Distributed Data

SDM kickoff meeting July 10-11, 2001
2
Task Grid I/O

• Goal Provide a NEW capability
• Provide PVFS under Grid FTP
• Coordinated with Globus group (ANL)
• Requires
• Globus source will be modified to work with PVFS
• PVFS extensions to be incorporated to optimize
  same node access (fast path access to PVFS
  stripes)
• Schedule
• Depends on two groups (Globus priorities)
• One year
• Evaluation
• Local fast path (PVFS)
• Across network and interface
• Action Item Bill will coordinate with Globus to
  determine use in Globus application (e.g.,
  GRYPHIN)

3
Grid I/O Plans

• MPI-IO front-end for the grid
• Selecting the right grid toolkit component
• Evaluate the capabilities of current MPI-IO hints
  mechanism under different scenarios (e.g., QoS)
• Extend and develop new hints (and mechanisms, if
  necessary) to enable Grid optimizations
• Schedule
• Year 2 and 3.

4
High-Dimensional Indexing Techniques

• Goal Reduce the amount of data needed to be
  transferred
• First Year
• Add string valued attribute oriented access
  capability
• On-line update Add new data without interrupting
  request processing
• Desired by STAR application
• People Involved
• John Wu (LBNL)
• Integrate with Area 1
• Contacts Pursue Astrophysics and Climate
  Applications

5
High-Dimensional Indexing

• Year 2 and 3 plans
• Investigate further reduction in data movement
  across Grid
• Investigate the use of MPI-IO capability to
  extract event level data
• Q. How do we translate event requests into MPI
  datatype requests? .