• INRIA PhD Student: Pierre Riteau
• INRIA Master Student: Djawida Dib

A cloud federation allowing live migration of virtual machines from one cloud to another would enable to dynamically adapt distributed applications, by relocating virtual machines to different clouds. This would not only enable migrating a virtual cluster from one cloud to another, but also relocating subsets of a virtual cluster. However, since network traffic between different clouds is subject to much higher latency than local network, this relocation needs to take into account communication patterns to limit communications crossing cloud boundaries.

In this work, we started the design and implementation of an autonomic adaptation system taking into account communication patterns of distributed applications. We developed and evaluated a subsystem that transparently captures network traffic from the VM and aggregates all statistics to infer communication patterns.