Power Log’n’Roll: Power-Efficient Localized Rollback for MPI Applications Using Message Logging Protocols

In fault tolerance for parallel and distributed systems, message logging protocols have played a prominent role in the last three decades. Such enable local rollback to provide recovery from fail-stop errors. Global techniques can be straightforward implement but at times lead slower than rollback. Local is more complicated offer faster times. this work, we study power energy efficiency implications of global We propose power-efficient version reduce consumption non-critical, blocked processes, using xmlns:xlink="http://www.w3.org/1999/xlink">Dynamic Voltage Frequency Scaling (DVFS) xmlns:xlink="http://www.w3.org/1999/xlink">clock modulation (CM). Our results 3 different MPI codes on 2 systems show that reduces CPU waste up 50% during phase, compared existing techniques, without introducing significant overheads. Furthermore, savings manifest all blocked which grow linearly with process count. estimate settings high overheads total reduced proposed