SCALE-EA: A Scalability Aware Performance Tuning Framework for OpenMP Applications

Abstract

HPC application developers, including OpenMP-based application developers, have stepped forward to endeavor the future design trends of exa-scale machines, such as, increased number of threads/cores, heterogeneous architectures, multiple levels of memories, and so forth; and, they have initiated procedures to address application level challenges, such as, data-driven scalability issues, energy consumption requirements, data availability needs, and so forth. Despite the existence of manual performance tuning solutions, users still deem it to be an intricate process. This paper proposes a scalability aware autotuning framework (SCALE-EA) that automatically identifies an efficient number of threads for OpenMP parallel regions using a Firefly Algorithm (FA) and a newly designed Modeling Assisted Firefly Algorithm (MAFA). MAFA of SCALE-EA was implemented in two approaches: Modeling Assisted Firefly Algorithm with Random Forest Modeling support (MAFA-RFM) and Modeling Assisted Firefly Algorithm with Linear Regression Modeling support (MAFA-LRM). The modeling and prediction algorithms of the proposed MAFA of SCALE-EA were based on the execution time and the hardware performance events of code regions of OpenMP applications. Experiments were conducted on two machines, namely, a Haswell based machine and an AMD Opteron based 48 core machine. The experimental results of the MAFA of SCALE-EA manifested the energy efficiencies of 31.21 to 77.3 percentage and the search time efficiencies of 5.53 to 32.56 percentage for candidate OpenMP applications such as CoMD, Arraybench, Taskbench, and Syncbench.