摘要
Power backup infrastructure is indispensable but its cost exceeds 20%of the whole power infrastructure expenditure of a traditional datacenter.Under-provisioning the power backup infrastructure has therefore been proposed to reduce the backup cost while still keep power available during grid power outages.However,how to under-provision the power backup facility for a green datacenter using renewable energy has not yet been explored.This paper proposes an approach to under-provision the power backup infrastructure as well as optimize the performance of individual applications with the presence of renewable energy,named Green-Up.It contributes two novel techniques:power source and power load managements.Power source management identifies the suitable power source for power backup whereas power load management selects the best power capping techniques for under-provisioning power backup infrastructure.In addition,we build hierarchical Bayesian Models to optimize power consumption and performance.To evaluate Green-Up,we build an experimental cluster consisting of 10 servers with a simulated solar power generator,and use four representative datacenter benchmarks.The experimental results demonstrate that our theoretical solutions provide a seamless bridge across the whole spectrum of renewable power availability and durations of grid power outages.
