On 6th of December I was in Shanghai presenting one of our last papers that we wrote at Barcelona Supercomputing Center for the 9th IEEE/ACM International Conference on Utility and Cloud Computing (UCC 2016).
Our current research work focuses on the area of Energy-aware Management for Modern Distributed Computing Systems. Our goal is to develop management algorithms for virtualized Data Centres in a large-scale distributed ecosystem running heterogeneous workloads that optimize their operation with respect to energy and ecological efficiency. The other co-authors of this particular paper are Mauro Canuto, Mario Macias and Jordi Guitart. The title of our paper is “A Methodology for Full-System Power Modeling in Heterogeneous Data Centers”. So essentially, we have developed a new power modeler that is compatible across multiple hardware platforms.
But why have we developed a new power modeler? That was to solve a series of limitations on current solutions that make power modelers application-dependent, platform-dependent, and sometimes inaccurate or excessively complex. In this paper, we propose a methodology that overcomes these limitations on power models and becomes platform and application-agnostic. What we do in this methodology is selecting systematically a minimum set of resource usage indicators and detect complex relations among them. At the end of this presentation we will show the results of our experiments. Our validation with real Cloud applications show that such models provide high accuracy (around 5% of average estimation error).
You can check out the rest of the slides at: Slideshare.