Energy-efficient runtime systems.

摘要

Improving energy efficiency is becoming imperative in today's computing as energy consumption may carry as much cost as the maintenance of computer systems. In fact, energy efficiency has become a first-class design goal with direct consequences on operational cost, reliability, usability, maintainability, and environmental sustainability. No matter how sophisticated a cooling system might get or how energy efficiency hardware might be, it always boils down to finding a software system that will extract maximum performance with minimal energy use. This dissertation describes the design, implementation, and evaluation of two innovative energy-efficient runtime systems for constructing energy-aware parallel programs.;First, this dissertation presents Hermes, an energy-efficient work-stealing language runtime. Work stealing is a promising approach to constructing multi-threaded program runtimes of parallel programming languages. The key insight is that threads in a work-stealing runtime environment---thieves and victims---have varying impacts on the overall program running time, and a coordination of their execution "tempos" can lead to energy efficiency with minimal performance loss. The centerpiece of Hermes is two complementary algorithms to coordinate thread tempos: the workpath-sensitive algorithm determines the tempo for each thread based on thief-victim relationships on the execution path, whereas the workload-sensitive algorithm selects the appropriate tempo based on the size of work-stealing deques. Hermes was constructed on top of Intel Cilk Plus' work-stealing runtime, and implements tempo adjustment through standard Dynamic Voltage and Frequency Scaling (DVFS).;Second, this dissertation presents AEQUITAS, an energy coordination system for multiple application-scoped energy-efficient runtimes. A growing number of energy optimization solutions operate at the application runtime level. Despite delivering promising results, these application-scoped optimizations are fundamentally greedy: they assume to have an exclusive access to power management "knobs". AEQUITAS is a first step to address this critical yet largely overlooked problem. The insight behind AEQUITAS is that co-existing applications view power-managing hardware as a shared resource and coordinate power management decisions. As a concrete instance of this philosophy, A EQUITAS was evaluated on top of Hermes' energy-efficient work-stealing runtime. Without AEQUITAS, multiple co-existing power-managing application runtimes tend to suffer significant performance losses while negating almost all power savings. With AEQUITAS, the beneficial energy-performance tradeoff reported in the single-application setting can be retained but in a much more challenging setting where multiple power-managing applications co-exist on parallel architectures and multiple CPU cores share the same power domain.;Experimental results show that both Hermes and AEQUITAS can lead to significant energy savings with minimal performance loss over commercial multi-core CPUs. Together, they provide a novel and systematic solution for optimizing energy efficiency of modern parallel language runtime systems.