Supervision : Pierre SENS

Co-supervision : MONNET Sébastien, KRISHNASWAMY Ruby

Designing Adaptive Replication Schemes for Efficient Content Delivery in Edge Networks

Content availability has become increasingly important for the Internet delivery chain. In order to enhance availability, content distribution network (CDN) providers have invested in hybrid designs that combines resources from datacenter and edge networks. Although, to deliver videos with an outstanding availability and meet the increasing user expectations, hybrid CDNs must enforce strict QoS metrics, like bitrate and latency, through SLA contracts. Adaptive content replication has been seen as a promising way to achieve this goal. However, it remains unclear how to avoid waste of resources when strict SLA contracts must be enforced.
In this dissertation, we focus on studying and evaluating adaptive replication schemes for a new generation of hybrid networks, whose resources come from consumers' devices, such as set-top boxes. To this end, we propose (i) Caju, a general-purpose content distribution system, which handles resource allocation in edge networks, and (ii) three novel adaptive replication schemes, AREN, Hermes, and WiseReplica. Extensive simulations with Caju show that our adaptive replication schemes are very efficient and can easily be extended to other CDN architectures. Finally, we provide some guidelines about our ongoing development of Caju in a world-wide testbed deployment.

2012-2021 Publications