Workload-Aware Runtime Energy Management for HPC Systems

摘要

Energy efficiency has become a crucial factor in high-performance computing, mainly due to its effect on operating costs and failure rates of computing platforms. To improve the energy efficiency of such systems, processors are equipped with low-power techniques such as dynamic voltage and frequency scaling (DVFS) and power capping. These techniques have to be controlled carefully as per the workload; otherwise, it may result in significant performance loss and/or power consumption due to system overheads (e.g. DVFS transition latency). Existing approaches are not effective in adapting to workload variations as they do not consider the combined effect of application compute-/memory-intensity, thread synchronization contention, and non?uniform memory accesses (NUMAs) owing to the underlying processor architecture. In this work, we propose a workload- aware runtime energy management technique that takes the aforementioned factors into account for efficient V-f control. The proposed technique measures the processor workload using Memory Accesses Per Micro-operation (MAPM), and also considers the thread synchronization contention and latency due to NUMAs to select an appropriate V-f setting. This approach also uses workload prediction for pro-active V-f control to improve the energy consumption and performance loss. The proposed technique has been implemented on the 12-core (24 threads) Intel Xeon E5-2630 and 61-core (244 threads) Xeon Phi many- core platforms, supporting per-core and system-wide DVFS, respectively. When evaluated with different application scenarios, results show an improvement in energy efficiency of up to 81.2% compared to existing approaches.