Modeling and optimization of survivable P2P multicasting

Various solutions based on Peer-to-Peer (P2P) multicasting have been gaining much popularity in recent years, since P2P multicasting can effectively support live streaming of various content. In this work we assume that the P2P multicasting is used to distribute content with high reliability requirements, e.g., weather warnings, security updates, financial data, security warnings, etc. The main idea to provide protection of the system against network failures is to establish several (at least two) disjoint multicasting trees. Our discussion in this paper centers on the problem how additional survivability constraints to provide failure-disjoint trees impact the operation of P2P multicasting systems. As the performance metrics we propose to use: streaming cost, maximum delay and throughput. The possible failure scenario we take into account is a single failure of one of the following network elements: streaming server, overlay link, uploading node and ISP link. We examine the topic of survivable P2P multicasting applying offline optimization methods and simulations. In the former case we formulate Mixed Integer Programming (MIP) models and use the CPLEX solver to obtain optimal results. For the streaming cost objective we compare two MIP formulations in terms of the complexity and execution time. Results show that our formulation provides much better performance compared to the classical P2P multicasting formulation proposed in the literature. Moreover, in the case of the streaming cost problem we propose a new evolutionary algorithm that yields results for larger networks than the CPLEX solver. The simulations are run to emulate a distributed network environment, in which each node makes its own decisions. Results obtained using both research methods confirm that the survivability of P2P multicasting can be achieved with relatively low additional system overhead for all three considered performance metrics: streaming cost, maximum delay and system throughput.