Multicore processors are becoming increasingly prevalent. As a result, there has been a dramatic surge of interest in parallel programming. Numerous ideas have been proposed over the last decade, spanning all disciplines of computer science, in an attempt to make it easier to write parallel programs: new programming models for parallelism, new synchronization constructs, software systems to dynamically parallelize applications, and hardware support for these models and systems. In short, a large number of solutions have been proposed to the problem of parallel programming.

What has been missing so far is a general strategy on how to parallelize programs. Prior work has largely focused on four approaches:

  • improving the performance of already parallel programs (such as those in the SPLASH-2 benchmark suite)
  • blindly applying parallelization techniques to sequential programs (such as those in the SPEC CPU2006 benchmark suite)
  • constructing synthetic microbenchmarks to demonstrate the utility of proposed parallelization techniques
  • developing parallelization techniques for specific operations (such as the matrix operations in the BLAS library)

None of these approaches answer the key question: how does a programmer parallelize a real-world sequential application in an efficient manner? To do so, we believe a programmer must understand (i) the algorithm the program implements, (ii) what kind of parallelism the program exhibits, (iii) what data structures are used in the program, and (iv) how those data structures affect the parallelization.

To help address this problem, we have released the Lonestar Benchmark Suite. It is a collection of widely-used real-world sequential applications that exhibit irregular behavior but still possess a significant amount of amorphous data parallelism. The Java and C++ applications in this suite are intended to serve as examples of data-parallel programs to which a programmer might apply various parallelization techniques.

If you would like to read more about or download the Lonestar Benchmark Suite, please use the links in the box to the right.


If you have questions or comments about the Lonestar Benchmark Suite, please contact us at lonestar@ices.utexas.edu.