Energy aware scheduling across #x2018;green#x2019; cloud data centres

摘要

Data centre energy costs are reduced when virtualisation is used as opposed to physical resource deployment to a volume sufficient to accommodate all application requests. Nonetheless, regardless of the resource provisioning approach, opportunities remain in the way in which they are made available and workload is scheduled. Cost incurred at a server is a function of its hardware characteristics. The objective of our approach is therefore to pack workload into servers, selected as a function of their cost to operate, to achieve (or as close to) the maximum recommended utilisation in a cost-efficient manner, avoiding instances where devices are under-utilised and management cost is incurred inefficiently. This is based on queuing theory principles and the relationship between packet arrival rate, service rate and response time, and recognises a similar exponential relationship between power cost and server utilisation to drive its intelligent selection for improved efficiency. There is a subsequent opportunity to power redundant devices off to exploit power savings through avoiding their management.