PI Name: Xingfu Wu, MCS

Description: In this project, we plan to explore and model the tradeoffs among time, power and resilience on heterogeneous systems (CPUs, GPUs, and FPGAs) of JLSE resources. We will develop an integrated tool to collect performance, power for different components such as nodes, CPUs, memory, disk, network, GPUs, and FPGAs, then use our MuMMI (Multiple Metrics Modeling Infrastructure) to build performance counter-based models for performance and power consumptions. Based on these performance and power models, we analyze the trade-offs for energy efficiency on the heterogeneous systems.

Testbed: Intel KNL, Skylake, Nvidia