Spring 05 Independent Study Midterm Review

1
Spring 05 Independent StudyMidterm Review

• Hans Sherburne
• HCS Research Laboratory
• University of Florida

2
Proposed Project

• Performance Analysis Tool Evaluation
• Purpose
• Obtain information regarding existing tools to
  assist in design of new tool for UPC/SHMEM
• Problem
• Evaluating performance is important, but
  inadequately supported for UPC/SHMEM
• Need for Solution
• Desire to learn from the mistakes and successes
  of others
• Benefits of Solution
• The information procured through evaluating
  existing tools will guide the development of a
  new one.

3
Accomplishments achieved while working with Adam

• Initially, I collaborated with Adam to develop a
  consistent approach to evaluating tools, while
  learning from his experience
• Testing Plan (1/10)
• Described general testing strategy
• Devised testing criteria for previously selected
  features
• PAT testing suite (1/20)
• Assembled a collection of programs to serve as
  input to tools being reviewed
• Overhead in tracing
• Effectiveness in analyzing various performance
  bottlenecks.
• Performance Analysis Tool List (1/21)
• Researched and assembled list of existing tools,
  and useful data
• Selected most interesting tools to evaluate
• MPE/Jumpshot Evaluation (1/25)
• Evaluated functionality of Jumpshot
  (visualization)
• Dimemas/Paraver/MPITrace (2/17)
• Evaluated functionality of Paraver
  (visualization)

4
Accomplishments achieved while working
independently

• Recently, Ive branched off on my own
• mpiP Evaluation (2/17)
• Evaluated the functionality of mpiP (profiling)
• Not originally scheduled for review, but included
  for its simplicity and effectiveness in display
  of statistical information
• Intel Trace Analyzer/Trace Collector Evaluation
  (3/15)
• Evaluation of the functionality of Trace
  Collector for data collection
• Evaluation of the functionality of Trace Analyzer
  for visual display.
• Will present this today!

5
Proposal Gantt Chart
6
Difficulties

• Learning many new skills and concepts, while
  putting them to use.
• Tool Evaluation
• It is tough to gauge how long an evaluation will
  take before doing it.
• MPI
• It is hard to judge the importance of some tool
  features without knowing how a real user would
  use it.
• UPC
• It is hard to judge how well a tool will extend
  to UPC without a good understanding of the
  innerworkings UPC.
• Programming/compiling C
• When a problem arises it is tough to determine if
  it is a problem related to the tool itself, or
  the environment.
• Linux
• The CLI is a powerful tool, but still fairly new
  to me.
• HCS lab environment
• Different clusters offer different functionality
• Some computer within clusters offer different
  functionality

7
For the rest of semester

• Tasks
• Evaluate SvPablo
• Assist with high-level prototype design
• Independent Study Report
• Goals
• Continue to learn more about MPI, and UPC (esp.
  UPC)
• Continue to learn more about working with Linux /
  C
• Continue to make valuable contributions to the
  UPC Group, and HCS Lab