Energy efficiency of P2P and distributed clouds networks

• Main Author: Lawey, Ahmed Qasim
• Other Authors: Elmirghani, Jaafar
• Published: University of Leeds 2015
• Subjects: 004.6
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.666625

Since its inception, the Internet witnessed two major approaches to communicate digital content to end users: peer to peer (P2P) and client/server (C/S) networks. Both approaches require high bandwidth and low latency physical underlying networks to meet the users’ escalating demands. Network operators typically have to overprovision their systems to guarantee acceptable quality of service (QoS) and availability while delivering content. However, more physical devices led to more ICT power consumption over the years. An effective approach to confront these challenges is to jointly optimise the energy consumption of content providers and transportation networks. This thesis proposes a number of energy efficient mechanisms to optimise BitTorrent based P2P networks and clouds based C/S content distribution over IP/WDM based core optical networks. For P2P systems, a mixed integer linear programming (MILP) optimisation, two heuristics and an experimental testbed are developed to minimise the power consumption of IP/WDM networks that deliver traffic generated by an overlay layer of homogeneous BitTorrent users. The approach optimises peers’ selection where the goal is to minimise IP/WDM network power consumption while maximising peers download rate. The results are compared to typical C/S systems. We also considered Heterogeneous BitTorrent peers and developed models that optimise P2P systems to compensate for different peers behaviour after finishing downloading. We investigated the impact of core network physical topology on the energy efficiency of BitTorrent systems. We also investigated the power consumption of Video on Demand (VoD) services using CDN, P2P and hybrid CDN-P2P architectures over IP/WDM networks and addressed content providers efforts to balance the load among their data centres. For cloud systems, a MILP and a heuristic were developed to minimise content delivery induced power consumption of both clouds and IP/WDM networks. This was done by optimally determining the number, location and internal capability in terms of servers, LAN and storage of each cloud, subject to daily traffic variation. Different replication schemes were studied revealing that replicating content into multiple clouds based on content popularity is the optimum approach with respect to energy. The model was extended to study Storage as a Service (StaaS). We also studied the problem of virtual machine placement in IP/WDM networks and showed that VM Slicing is the best approach compared to migration and replication schemes to minimise energy. Finally, we have investigated the utilisation of renewable energy sources represented by solar cells and wind farms in BitTorrent networks and content delivery clouds, respectively. Comprehensive modelling and simulation as well as experimental demonstration were developed, leading to key contributions in the field of energy efficient telecommunications.