LSU EE 7700-3 -- Advanced Computer Implementation -- Spring 2005

Lecture Set 2

Research Project Steps,  Research Paper Sections

====================


 Idea
   Come up with idea.
   E: Why not put jump targets in GHR

 Scrutinize Idea
   Potential for improvement?
   Already done?
   Make go/nogo decision.

 Analyzing Potential
   Find normal/ideal performance data. (E.g., Wall limit study.)
   Estimate how often idea would deliver improvements.
   Estimate how often idea would worsen performance!
   Perhaps quickly collect data for this.

 Already Done?
   Ask.
   Find recent papers on related area, look in "Prior Work" section.

 Evaluate
   Modify simulator to implement technique.
   Modify simulator to implement comparison techniques.
   Run simulation: qsub -q 128  myidea
   Tune as necessary.

 Simulation Data
   Performance of base (normal, conventional) system.
   Performance with idea.
     Is it worth it?
   Potential of idea, how much of potential is reached.
     Can it be improved further?
   Parameter sensitivity.
     Is it really worth it?

 Example Data
   Prediction accuracy of branch after (what looks like) switch statement.
   Number of such branches compared to total number of branches.

 Publish
   If speedup < 10% repeat earlier step(s)

Paper Organization

 Abstract
   -- I don't have time to read this, what are the main ideas?
   Summary of paper, NOT an introduction.
   Put in major ideas, results, and conclusions. 
   Yes, ruin the surprise ending: Rosebud is a XXXXXX!!!!

 Introduction
   -- What problem was solved and how, what's in the rest of the paper.
   Quickly describe problem.
   Present key ideas used to solve it.
   Tick off benefits, relation to prior work, etc.
   Summarize the rest of the manuscript.

 Preliminaries
   Boring definitions. Keep it short.  This section not always present.

 Mechanism
   -- Exactly how does it work?
   How does it work.
   Don't go into too much detail.

 Methodology
   -- Was the idea tested realistically?
   Simulator and configuration.

 Results
   -- Does it really work?
   Show performance.
   Show that it works for the right reasons.

 Prior Work
   -- What does it build upon?
   Describe other approaches to same problem.
   Differentiate the idea.

 Conclusions
   -- Did I miss anything?
   I'm not good at this part.