Dynamic execution platforms over federated clouds

The increasing needs for computing power have led to parallel and distributed computing, which harness the power of large computing infrastructures in a concurrent manner. Recently, virtualization technologies have increased in popularity, thanks to hypervisors improvements, the shift to multi-core architectures, and the spread of Internet services. This has led to the emergence of cloud computing, a paradigm offering computing resources in an elastic, on-demand approach while charging only for consumed resources. In this context, this thesis proposes four contributions to leverage the power of multiple clouds. They follow two directions: the creation of elastic execution platforms on top of federated clouds, and inter-cloud live migration for using them in a dynamic manner. We propose mechanisms to efficiently build elastic execution platforms on top of multiple clouds using the sky computing federation approach. Resilin is a system for creating and managing MapReduce execution platforms on top of federated clouds, allowing to easily execute MapReduce computations without interacting with low level cloud interfaces. We propose mechanisms to reconfigure virtual network infrastructures in the presence of inter-cloud live migration, implemented in the ViNe virtual network from University of Florida. Finally, Shrinker is a live migration protocol improving the migration of virtual clusters over wide area networks by eliminating duplicated data between virtual machines.