Execution Traces and Data Redistribution

1
Execution Traces and Data Redistribution

• Presented By
• Sandip Tikar

2
Data Grid

• Data Grid Motivations.
• Replica Management
• - For better performance or availability to
  accesses
• - Mainly for access to published
  resources read-only model
• Chimera A Virtual Data System

3
Execution Traces

• What is Execution Trace ?
• - Nothing but a log of problem run.
• Why it is needed ?
• - For Numerical Reproducibility.

4
(No Transcript)
5
Execution Traces in GradSolve

• GradSolve a RPC System.
• Staging of input data.
• Storing of Execution Trace.
• Using Execution Trace.
• User has to specify explicitly whenever he want
  to use
• Execution Traces.

6
Our Approach

• Automatic Discovery.
• Storage of Execution Traces
• - Validity.
• - How many traces to be saved for a
  problem?
• - Cleanup Routine.
• What if Same problem runs with different Data
  Distribution ?

7
Data Redistribution

• How to aggregate different data segments from
  different data distributions to form new
  distribution?
• Cost analysis for aggregation of data.

8
References

• GradSolve RPC for High Performance Computing on
  the Grid
• S. Vadhiyar, Jack Dongarra.
• A Service-Oriented Grid Computing Platform and
  Its Framewor
• Wei Jie, Tianyi Zang, Terence Hung Zhou Lei
• Replica Selection in the Globus Data Grid
• Sudharshan vazhkudai, Steven Tuecke, Ian
  Foster
• Chimera A Virtual Data System for Representing,
  Querying and Automating Data Derivation.
• Ian Foster, Jans Vockler , Michael Wilde ,
  Yong Zhao.