DiBA: Distributed Power Budget Allocation for Large-Scale Computing Clusters

Power management has become a central issue in large-scale computing clusters where a considerable amount of energy is consumed and a large operational cost is incurred annually. Traditional power management techniques have a centralized design that creates challenges for scalability of computing clusters. In this work, we develop a framework for distributed power budget allocation that maximizes the utility of computing nodes subject to a total power budget constraint. To eliminate the role of central coordinator in the primal-dual technique, we propose a distributed power budget allocation algorithm (DiBA) which maximizes the combined performance of a cluster subject to a power budget constraint in a distributed fashion. Specifically, DiBA is a consensus-based algorithm in which each server determines its optimal power consumption locally by communicating its state with neighbors (connected nodes) in a cluster. We characterize a synchronous primal-dual technique to obtain a benchmark for comparison with the distributed algorithm that we propose. We demonstrate numerically that DiBA is a scalable algorithm that outperforms the conventional primal-dual method on large scale clusters in terms of convergence time. Further, DiBA eliminates the communication bottleneck in the primal-dual method. We thoroughly evaluate the characteristics of DiBA through simulations of large-scale clusters. Furthermore, we provide results from a proof-of-concept implementation on a real experimental cluster.