From Idea to Proposal

• START SAVING REFERECNES NOW.
  • We will discuss this later, but ...
  • I use bibtex, so I export this citation.
  • I also save a PDF copy of the file whenever I can.
• Most of us have met and have anything from a beginning idea to a semi developed topic.
  • At the time of writing, those of you who have not yet met with me have lost 1% of your final grade.
  • And you are behind in your work.
• So what do we do now?
  • Narrow the topic!
  • Many topics you find can be a lifetime of work.
  • You need to find something that is manageable for this project.
• Let's take a look at Jack Dongarra
  • He is huge in the parallel world.
  • Look at his publication list by date: The most current topics are very
  • Look by cited by
    • These are general publications.
    • But he IS the expert in these systems.
  • We can't be experts yet. We need to start somewhere more restricted.
• An example Search
  • A word of caution, I spend a LONG time making these notes.
    • Like when I program in class, I HOPE that this looks like I am competent.
    • But you know there were MANY errors involved in the process.
    • and much backtracking
    • Research, like writing a difficult program is messy and error prone.
    • Like anything new or different you do in life.
  • I am going to start looking at MPI.
  • Go to scholar.google.com
    • I added Baron-Forness library under settings.
  • Search MPI
    • Happy Day About 1,040,000 results
  • This seems a bit too wide.
  • At this point I really should go to my inspiration
    • Characterization of MPI Usage on a Production Supercomputer
    • But we want to work on narrowing the topic.
  • So back to scholar
    • Why scholar and not one of the library databases?
  • Search for MPI performance
    • 300,000 results. Argh.
    • Limiting by time might be temping, but I worry about that a bit.
      • Some of our topics are older and a time limit could be harmful.
      • And sometimes you want a "foundational" article.
      • Discuss this.
    • Note scholar does present some suggestions in Related Searches
      • I need a project so I need something I can do.
      • I like the mpi performance profiling that showed up for me.
      • That trimmed it down to 25,000 results.
      • Note the related searches are also more detailed.
  • Right now I am looking for
    • Something I like.
    • Something Newer
    • Something with quite a Cited By links.
    • So I am going to time limit this
  • Limit to 2021
    • 1,200 results.
    • But as I scan the list I see
      • Not cited by anyone (this is understandable)
      • Really specific topics (this is good)
      • A better related search suggestion list.
  • I've mentioned cited by many times, why is citation count important?
  • I clicked on automatic profiling of MPI applications
    • This cleared the date but two articles pop up here that catch my eye
      • Automatic MPI counter profiling of all users: First results on a CRAY T3E 900-512, 101 citations but published in 1999
      • Dynamic software testing of MPI applications with Umpire, 2000 cited by 238
      • These two
        • are very old
  • I picked the UMPIRE article, select citations and limited to 2018 and later
    • 38 results. Worth a look.
    • I like a number of these papers.
      • Detection of Bugs in MPI Programs Using a Hybrid Approach
        • This is a master's thesis
      • Genetic algorithm based test data generation for MPI parallel programs with blocking communication
      • COMPI: Concolic Testing for MPI Applications
  • What to do?
    • Read the abstracts, look at the conclusions.
    • Am I interested in investing time in this subject?
    • Are there any interesting questions in the Conclusions?
    • Are there significant references?
  • Let's look at the Genetic algorithm based test data generation for MPI parallel programs with blocking communication article.
    • Note there are some VERY OLD articles
      • These are probably foundational articles, and good for a forward search like we have just done.
      • Actually the fact that there are not a lot of new articles tells me in the bib
        • There is not a lot of related work (good subject, new subject, really novel approach)
        • Or the authors are not staying current with their bib search.
        • Or the journal publication time killed them.
        • Look at the footnote for Fei-Ching Kuo, they passed in 2017.
          • The paper has been cited by at least 10 other articles already.
          • Go back to scholar and there are MANY related articles.
          • I think this is a good paper.
      • What next:
        • Read the article in detail. This may take some time and effort.
        • Look at the citing articles, are some of them interesting as well
        • Look at the bib, any articles there interesting.
        • Look at the authors, (Use google scholar or find their personal web pages.)
          • They most likely have published related articles.
        • Look on Science direct as well.
      • Can I turn this into a project?
        • Can I form a question?
        • Debugging MPI programs - too wide
        • Test sets for debugging MPI programs - still too wide
        • Can I reproduce the results presented in Genetic algorithm based test data generation for MPI parallel programs with blocking communication - ok, probably good for this class, but perhaps a Little limited.
    • What do you need to do
      • Find a foundational article in an area or
      • Find an article you like
      • Trace the bib tree backwards or forwards to find a "question".
    • Start building a bib, or at least a list of references.