A framework for monitoring energy consumption in multicore architectures

Modern processors provide a specialized interface of services to access data describing the energy consumption at runtime. However, such interfaces are not standardized nor they perform any data treatment or uniformization. This is undesirable to the end programmer and reduces the portability of solutions that are built upon that kind of information. This paper presents a framework that describes a common interface of services that expose processors' energy consumption. The developed solution has a modular and extensible architecture design and provides a uniform data representation for the collected data. In this approach, energy consumption of parallel programs running on a multicore-based architecture was monitored dynamically, independently from the processor family supporting the execution. Our work was validated through two case of studies that claim its applicability.