Shafik, Rishad Ahmed, Das, Anup K., Yang, Sheng, Merrett, Geoff V. and Al-Hashimi, Bashir
Learning-based runtime management of energy-efficient and reliable many-core systems
At ESWEEK Workshop on Compiler Assisted SoC Assembly (CASA), India.
This paper highlights and demonstrates our research works to date addressing the energy-efficiency and reliability challenges of many-core systems through intelligent runtime management algorithms. The algorithms are implemented through cross-layer interactions between the three layers: application, runtime and hardware, forming our core theme of working together. The annotated application tasks communicate the performance, energy or reliability requirements to the runtime. With such requirements, the runtime exercises the hardware through various control knobs and gets the feedback of these controls through the performance monitors. The aim is to learn the best possible hardware controls during runtime to achieve energy-efficiency and improved reliability, while meeting the specified application requirements.
Actions (login required)