next up previous contents index
Next: Putting the Experiment under Up: A Tool Set for Previous: Installation   Contents   Index


Making Experiments Reproducible

A typical scenario in which the Tool Set for Computational Experiments might be of great use is that we have a program that performs certain experiments and we want to be able to repeat or at least recapitulate later (possibly after several years) how these experiments were done.


Generally, the following two steps are necessary to make an experiment reproducible.

  1. All source files that are necessary to perform the experiment are archived.
  2. Certain parameters used for the compilation and execution of the experiment are logged.


The easiest way to do this is to use the Tool Set for Computational Experiments in combination with a version control system such as the Concurrent Version System (CVS) (available at http://www.cvshome.org/). Using CVS enables us to restore all source files of a particular experiment if we know, for example, the date on which the experiment was run. The bookkeeping mechanisms provided by the tool set (in particular, the program labrun) record the dates of experiments as well as many other relevant environmental data, which allows one later to rerun the experiment (provided that the same environment, i.e., machine, compiler, etc., is still available). Rerunning is easily accomplished using the labrerun tool.



Subsections
next up previous contents index
Next: Putting the Experiment under Up: A Tool Set for Previous: Installation   Contents   Index
Tobias Polzin 2002-11-18