Empirically Examining the Parallelizability of Open Source Software System

An empirical study is presented that examines the potential to automatically parallelism, using refactorings tools and/or compilers, 11 open source software. Static analysis methods are applied to each system to determine the number of for-loops and free-loops (i.e., loops that can be parallized). For each non-free loop the various inhibitors (to parallelization) are determined and counted. The results show that function calls within for-loops represent the vast majority of inhibitors and thus pose the greatest roadblock to adapt and re-engineer systems to better utilize parallelization. This is somewhat contradictory to the literature, which is focused primarily on the removal of data dependencies within loops. Additionally, the historical data of inhibitor counts for the set of systems is presented over a ten-year period. The data shows few of the systems examined are increasing the potential to parallelizable loops over time.