LPCLOUD: Optimum management of low-power modes for cloud computing


Optimum management of low-power modes for cloud computing

Funding: National Programme for Public-Private Cooperation, INNPACTO (MINECO)
Partners: Ermestel, Elite TC, CeSViMa (UPM), LSI (UPM)
Duration: 2013-2015
Principal Investigator: Alberto Carrillo (Ermestel)
UPM-LSI Manager: José M. Moya

Low-power modes (Dynamic Voltage and Frequency Scaling, DVFS) have had a major impact in reducing consumption of desktop computers. It is estimated that these techniques save up to 70% of energy consumption in these machines.
However, until now, frequency scaling and power supply have minor impact on the data center due to the associated performance loss, when switching between DVFS modes. For this reason the only thing that is commonly used is shutting down unused machines in cloud computing environments.

Although in dedicated data centers, due to the high utilization of resources, the impact of not using low-power modes is not as important, in cloud computing, where demand is highly variable and the average utilization of resources is quite low, appropriate use of management techniques for low-power modes could reduce energy consumption without any major performance penalty.

The overall objective of this project is to develop techniques for hardware-software global management (data center level) of the low-power modes of computers, taking into account the characteristics of applications, energy demand, performance requirements, previous history of execution in the data center, the characteristics of computing and cooling resources, and the current data center status.