ABSTRACT

This chapter deals with power management in high-performance computing (HPC) systems. Power consumption of battery operated devices has been an issue for a long time. Over the last decade power consumption of HPC systems has emerged as a significant problem that even limits future development. Striving for performance has resulted in an enormously high peak power draw. The struggle for performance is reflected in the Top500 list [21] of the 500 most powerful supercomputers. The number one ranked system of the June 2010 list, Jaguar [20], comprises an incredible number of almost 225,000 cores and it brings the theoretical peak capability to 2.3 petaflop/s. Jaguar requires almost 2.8 times the electric power of its predecessor Roadrunner, the number two on the same list. This difference translates into millions of dollars per year in operating costs. Power estimates for exascale computer power dissipation range from many tens to low hundreds of megawatts [14]. Hence power consumption is one of the most important design constraints for HPC centers nowadays. Besides a tremendous increase in cost of ownership, power-awareness in HPC centers is motivated by other reasons such as system reliability and environmental footprint. Recently, the Top500 list has been accompanied by the Green500 list ranking the most energy efficient supercomputers in FLOPS/w.