Markov approximation for combinatorial network optimization
INFOCOM'10 Proceedings of the 29th conference on Information communications
| Hi-index | 0.00 |
We study the problem of neighbor selection in BitTorrent-like peer-to-peer (P2P) systems, and propose a "soft-worst-neighbor-choking" algorithm that is provably optimal. In practical P2P systems, peers often keep a large set of potential neighbors, but only simultaneously upload/download to/from a small subset of them, which we call active neighbors, to avoid excessive connection overhead. A natural question to ask is: which active neighbor set should each peer choose to maximize the global system performance? The combinatorial nature of the problem makes it especially challenging. In this paper, we formulate an optimization problem and derive a distributed algorithm. We remark that our solution has a similar favor compared to the worst neighbor choking and optimistic unchoking neighbor selection algorithms that are implemented by BitTorrent. However, it encourages peers to stick to better performing neighbors for longer time and is provably globally optimal. Our proposed solution is easy to implement: each peer periodically waits for a constant period of time that depends on the size of the potential neighbor set and the aggregated utility of the active neighbors, chokes (drops) one of its current active neighbors with probability proportional to an exponential weight on the utility of the corresponding link, and randomly unchokes (adds) a new neighbor from its potential neighbor set. Our theoretical findings provide insightful guidelines to designing practical P2P systems. Simulation results corroborate our proposed solution.