Task Bench: A Parameterized Benchmark for Evaluating Parallel Runtime Performance

Authors: Elliott Slaughter (SLAC National Accelerator Laboratory); Wei Wu (Los Alamos National Laboratory); Yuankun Fu (Purdue University); Legend Brandenburg, Nicolai Garcia, Wilhem Kautz, Emily Marx, and Kaleb S. Morris (Stanford University); Qinglei Cao and George Bosilca (University of Tennessee); Seema Mirchandaney (SLAC National Accelerator Laboratory); Wonchan Lee and Sean Treichler (Nvidia Corporation); Patrick McCormick (Los Alamos National Laboratory); and Alex Aiken (Stanford University)

Abstract: We present Task Bench, a parameterized benchmark designed to explore the performance of distributed programming systems under a variety of application scenarios. Task Bench dramatically lowers the barrier to benchmarking and comparing multiple programming systems by making the implementation for a given system orthogonal to the benchmarks themselves: every benchmark constructed with Task Bench runs on every Task Bench implementation. Furthermore, Task Bench’s parameterization enables a wide variety of benchmark scenarios that distill the key characteristics of larger applications.

To assess the effectiveness and overheads of the tested systems, we introduce a novel metric; minimum effective task granularity (METG). We conduct a comprehensive study with 15 programming systems on up to 256 Haswell nodes of the Cori supercomputer. Running at scale, 100μs-long tasks are the finest granularity that any system runs efficiently with current technologies. We also study each system’s scalability and ability to hide communication, and mitigate load imbalance.




Back to Technical Papers Archive Listing