CS 194 Project Checkpoint

1
CS 194 Project Checkpoint

• Exploring Latency Constraints of Co-Processing
  Boards
• Grant Jenks
• UCLA

2
Outline

• Original Plan
• What I had planned to do.
• Accomplishments
• What I actually did.
• Problems
• What happened.
• Revised Plan
• Whats the new plan.

3
Original Plan

• Analyze and Synthesize
• First quarter Analysis
• Develop the test platform.
• Test the different classes of programs on the
  platform and collect a lot of data.
• Second quarter Synthesis
• From the data, determine that sweet spot.
• Based on analysis, determine a way to validate
  the initial tests using existing hardware or
  FPGAs.
• If validation simply cannot happen then more
  fully develop why this is so.

4
Accomplishments

• Vtune Tests
• Dozens of different processor loads recorded.
• This data can demonstrate a strong need for
  another processor.
• Classic PTLsim Tests (32-bit)
• Dozens of different programs with various
  parameters were run to determine optimal
  performance in a one processor and one program
  environment.
• This data can demonstrate the optimal performance
  of a program with no contention for memory or
  processing.
• Analyzed code and evaluated how to change it so
  that the proper simulation environment can be
  created.
• Full Simulation PTLsim/XEN (64-bit)
• Modified kernel code and patched kernels to allow
  for PTLsim enhancements.
• Gained lots of experience in systems
  administration tasks and operating systems
  practicals.
• Tried dozens of different configurations trying
  to get this to work.

5
Problems Faced

• Hardware Issues
• Havent been able to acquire a 64-bit computer
  with virtualization technology support.
• SATA Hard Drive failed in test machine.
• PTLsim/XEN
• Code is difficult to change.
• Systems work requires tons of time while juggling
  operating systems.
• XEN technologies are still fairly new and not
  well supported.
• Little help in general with XEN technologies.
• Vtune
• Reproducibility is sometimes challenging.
• Bad Luck
• Nothing works out-of-the-box
• Scripts have been tweaked to make builds and runs
  work properly.

6
Revised Plan

• Produce working simulation environment under
  PTLsim/XEN.
• Run a wide variety of tests in the simulation
  environment under different hardware scenarios.
• The goal is to get at least five different
  software scenarios tested under at least five
  different hardware simulations.
• Determine the sweet spot for a hardware
  implementation of a co-processing board.
• Go back and run more tests in Classic PTLsim or
  Vtune if necessary.
• Time permitting, validate sweet spot hardware
  implementation with real-world co-processing
  board.

7
Summary

• Original plan did not properly estimate the time
  it would take to set up the simulation
  environment.
• Data collected can establish the need for more
  processing power and show optimal program
  execution.
• Theres a lot of momentum going into the second
  quarter but it is questionable whether the tests
  can be verified.