CHOOSING: Cooperation on Hybrid cOmputing clOuds for energy SavING
This project is a collaborative project between several centers (LIG,LIP6, LRI, UNICAMP, USP).
A brief description
The cloud computing is an important factor for environmentally sustain- able development. If, in the one hand, the increasing demand of users drive the creation of large datacenters, in the other hand, cloud computing’s “mul- titenancy” trait allows the reduction of physical hardware and, therefore, the saving of energy. Thus, it is imperative to optimize the energy consumption corresponding to the datacenter’s activities. Three elements are crucial on energy consumption of a cloud platform: computation (processing), storage and network infrastructure. Therefore, the aim of this project is to provide different techniques to reduce energy con- sumption regarding these three elements. Our work mainly focuses on energy saving aspects based on virtualization, i.e., pursuing the idea of the intensive migration of classical storage/processing systems to virtual ones. We will study how different organizations (whose resources are combined as hybrid clouds) can cooperate with each other in order to minimize the energy consumption without the detriment of client requirements or quality of service. Then, we intend to propose efficient al- gorithmic solutions and design new coordination mechanisms that incentive cloud providers to collaborate.
Project Meeting, Wednesday, March 30, 2016, Aussois
-
Controlling and optimizing energy
- Speaker : David Gleiser
-
A Runtime for Code Offloading on Modern Heteregeneous Platforms
- Speaker : Rogério Goncalves
- Abstract : The runtime part of framework supports the code offloading based on code versioning of parallel loops.The idea is that the OpenMP input code can be generated by compiler or written manually. The decision about offloading has been taken automatically at runtime using the operational intensity. The runtime libraries are intercepting some applications calls for OpenMP runtime using a hooking technique. They are collecting measures in all levels of memory hierarchy and the data transfers between host and devices have been considered on decision.
-
Hybrid scheduling, approximation ratio and energy.
- Speaker : Grégory Mounié
- Abstract : Some scheduling algorithms use the dual approximation scheme to build a solution with a performance guarantee. In addition to minimize the makespan, such algorithms can simultaneously take into account another resource usage. For example, They can trade off makespan and energy consumption, as energy is directly related to the number of used computing resources. We present an algorithm for scheduling independent tasks on \(m\) CPU and \(k\) GPU. The algorithm builds a family of approximated Pareto solutions, minimizing \(m\), or \(k\), for a chosen targeted makespan.
-
Title : Probabilistic Byzantine Tolerance for Cloud Computing
- Speaker : Luciana Arantes
- Abstract : This work that extends the former by proposing a probabilistic byzantine tolerance scheduling for hybrid clouds. This a cooperation work with Roy Freedam from Technion University, Israel.,
Participants of this project
Luciana Arantes (LIP6), Evripidis Bampis (LIP6), Daniel Batista (USP), Luiz Bittencourt (UNICAMP), Johanne Cohen (LRI), Daniel Cordeiro (USP), Pierre-François Dutot (LIG) , Alfredo Goldman (USP), Rogério Goncalves (LIG), Jean-Francois Mehaut (LIG), Grégory Mounié (LIG), Joanna Tomasik (LRI), Denis Trystram (LIG), and Raphael Yokoingawa de Camargo (USP),