摘要
In this research, we explore the defragmentation of allocated compute resources so as to conserve energy on an IBM Blue Gene/Q. We examine a real trace from a new four-rack system and explore through simulation three heuristics to minimise energy waste through defragmentation. We describe a number of heuristics for detecting when it is desirable from an energy standpoint to defragment the computing resource through checkpoint/restart. Using heuristics, we were able to gain a simulated saving of 4.36 % of total system power. When applied to all BlueGene/Qs on the Top500 list, this is the equivalent of running the average US household for 698.5 years per annum.