Abstract—Massive content distribution on overlay networks stresses both the server and the network resources because of the large volumes of data, relatively high bandwidth requirement, and many concurrent clients. While the server limitation can be circumvented by replicating the data at more nodes, the network limitation is far less easy to cope with, due to the difficulty in determining the cause and location of congestion and in provisioning extra resources. In this paper, we present novel schemes for massive content distribution, that assign the clients to appropriate servers, so that the network load is reduced and also well balanced, and the network resource consumption is low. Our schemes allow scaling to very large system because the algorithms are very efficient and do not require network measurement, or topology or routing information. The core problems are formulated as partitioning the clients into disjoint subsets according to the degree of interference criterion, which reflects network resource usage and the interference among the concurrent connections. We prove that these problems are NP-complete, and present heuristic algorithms for them. Through simulation, we show that the algorithms are simple yet effective in achieving the design goals.

In this paper, we investigate the issue of node selection for parallel access in overlay networks, which is a fundamental problem in nearly all recent content distribution systems, grid computing or other peer-to-peer applications. To achieve high performance and resilience to failures, a client can make connections with multiple servers simultaneously and receive different portions of the data from the servers in parallel. However, selecting the best set of servers from the set of all candidate nodes is not a straightforward task, and the obtained performance can vary dramatically depending on the selection result. In this paper, we present a node selection scheme in a hypercube-like overlay network that generates the optimal server set with respect to the worst-case link stress (WLS) criterion. The algorithm allows scaling to very large system because it is very efficient and does not require network measurement or collection of topology or routing information. It has performance advantages in a number of areas, particularly against the random selection scheme. First, it minimizes the level of congestion at the bottleneck link. This is equivalent to maximizing the achievable throughput. Second, it consumes less network resources in terms of the total number of links used and the total bandwidth usage. Third, it leads to low average round-trip time to selected servers, hence, allowing nearby nodes to exchange more data, an objective sought by many content distribution systems.