Title :
Star-Cap: Cluster Power Management Using Software-Only Models
Author :
Davis, John D. ; Rivoire, Suzanne ; Goldszmidt, Moises
Abstract :
Star-Cap is a high-fidelity, real-time power management system for clusters. Star-Cap´s cluster power management module uses software-only power models to implement a proactive power capping mechanism with the ability to distribute a cluster´s power budget non-uniformly across nodes. We evaluated Star-Cap on a variety of MapReduce-style workloads running on a low-power cluster. Depending on application and platform, Star-Cap improves cluster throughput by 14-43% compared to using a uniform power cap distribution policy with a purely reactive power capping mechanism, even one based on real physical measurements. Furthermore, Star-Cap maintains cluster throughput while reducing overall cluster power budget by 14% with no additional capital or operating cost. Star-Cap´s proactive power capping mechanism also improves response time to power cap violations by a factor of 2 to 10. Star-Cap´s overhead for collecting, computing, and reporting data is less than 1% CPU utilization.
Keywords :
computer centres; power aware computing; MapReduce-style workload; Star-Cap system; cluster power management system; power cap distribution policy; proactive power capping mechanism; purely reactive power capping mechanism; software-only model; Power demand; Power measurement; Predictive models; Radiation detectors; Servers; Software; Software algorithms;
Conference_Titel :
Parallel Processing Workshops (ICCPW), 2014 43rd International Conference on
DOI :
10.1109/ICPPW.2014.27