Network-Aware Epidemic Broadcast

Epidemic multicast is an emerging resilient and scalable approach to the reliable dissemination of application data in the context of very large scale distributed systems. Unfortunately, the resilience and scalability come at the cost of considerable redundancy which led to high resource consumption on both links and nodes. In environments with resource constrained links, such as in Cloud Computing infrastructure composed by data centers organized in a federation around the globe, the high resource consumption precludes the use of this class of protocols. The goal of this dissertation is therefore to cope with the constraints of these scenarios, by reducing the network load imposed on the constrained long distance links. This is achieved by constructing an overlay that re ects the characteristics of the links, and by using a dissemination protocol that takes into account locality when transmitting the message payloads. We conducted an extensive experimental evaluation that presents an improvement over an order of magnitude in the number of messages that traverse the costlier links, without endangering the resilience and scalability properties that make epidemic based protocols so attractive.