A Green Distributed Cooperation for Network and Content Management
MetadataShow full item record
CitationChiaraviglio, Luca; Matta, Ibrahim. "A Green Distributed Cooperation for Network and Content Management", Technical Report BUCS-TR-2010-008, Computer Science Department, Boston University, March 31, 2010. [Available from: http://hdl.handle.net/2144/3787]
We propose a distributed approach in which an Internet Service Provider (ISP) and a Content Provider (CP) cooperate to minimize total power consumption. Our solution is distributed between the ISP and the CP to limit shared information, such as network topology and servers load. In particular, we develop different algorithms adopting dual decomposition and Benders decomposition techniques. We investigate the performance of the proposed solutions on realistic case-studies. We compare our algorithms with a centralized model, whose aim is to minimize total power consumption. We first adopt convex functions to model power consumption of devices: all the distributed algorithms find optimal solutions in this scenario. We then introduce the possibility of powering off devices. Results show that in this case the distributed algorithms are close to the optimal solution, with a power efficiency loss less than 18%. For the proposed algorithms we speculate on the trade-off between the complexity of cooperation and that of the implementation. In particular, with the dual decomposition approach only the Lagrange multipliers associated with the traffic demands and users delay are shared between the ISP and CP, but a real implementation requires a trusted third-party server and careful tuning of parameters. On the contrary, with a Benders decomposition technique both the traffic demands and the ISP power consumption need to be shared, but this information is exchanged directly. Moreover, the parameters are easy to set, but the computational time grows linearly with the number of iterations. Finally, we investigate improvements to balance the power savings between the ISP and the CP.