Load balancing framework comparison

The main goal of this paper is to show how the load balancing frameworks Lace, OpenMP, Intel Thread Building Blocks (TBB) and CilkPlus perform compared to each other. The research focuses on the task-based nature of these frameworks, looking at the overhead of task creation but also at the overall performance of the frameworks. Tests were run using three benchmarks: Unbalanced Tree Search, Fibonacci and Strassen matrix multiplication. The benchmarks were compiled with the Intel Compiler as well as the GNU Compiler to study what consequences this has for the performance. The experiments revealed that out of the four frameworks Lace performs best, CilkPlus and TBB intermediate while OpenMP had the worst performance by far. CilkPlus was a little bit better than TBB. Promising options for future work are discussed.