Achieving Energy Eciency During Collective Communications
Sundriyal, Vaibhav and Zhang, Zhao and Sosonkina, Masha (2011) Achieving Energy Eciency During Collective Communications. Publisher UNSPECIFIED.
Full text available as:
Energy consumption has become a major design constraint in modern computing systems. With the advent of peta ops architectures, power-ecient software stacks have become imperative for scalability. Techniques, such as dynamic voltage and frequency scaling (called DVFS) and CPU clock modulation (called throttling) are often used to reduce the power consumption of the compute nodes. To avoid signicant performance losses, these techniques should be used judiciously during parallel application execution. For example, its communication phases may be good candidates to apply the DVFS and CPU throttling without incurring a considerable performance loss. They are often considered as indivisible operations while little attention is being devoted to the energy saving potential of their algorithmic steps. In this work, two important collective communication operations, all-to-all and allgather are investigated as to their augmentation with energy saving strategies on the per-call basis. The experiments prove the viability of such a ne-grain approach. While keeping the performance loss low, the obtained energy savings were always signicantly higher than those achieved when DVFS or throttling were switched on across the entire application run.
Archive Staff Only: edit this record