数据中心间空闲带宽感知的内容分发算法

Leftover bandwidth-aware peer selection algorithm for inter-datacenter content distribution

针对数据中心链路上存在时间窗不重叠的空闲带宽的情况,提出了利用该带宽分发容迟数据的基本思路,进而设计了一种分布式可扩展的空闲带宽感知的节点选择算法LBAPS,该算法避免了集中优化,适合目标节点较多的情况。为了匹配最优的带宽空闲节点,LBAPS按综合度量进行节点选择;为了优先把文件块上传到空闲带宽大的节点以及尽早把不同的块分布到更多节点,LBAPS按阈值预留资源以及按时间片退出上传。基于LBAPS实现了内容云原型系统P2PStitcher。PlanetLab上的实验表明,LBAPS算法所提出的策略可以有效地减少平均分发时间。

Due to the fact that leftover bandwidth appears during non-overlapping time intervals, an approach of using such bandwidth to distribute delay tolerant data was proposed, and then a distributs and scalable leftover band- width-aware peer selection algorithm named LBAPS was designed. LBAPS avoids centralized optimization method that fails to effectively utilize leftover bandwidth when multiple destinations occur. In LBAPS, a node selection strategy based on synthetical evaluation was presented in order to find appropriate nodes with leftover bandwidth currently. In addition, two other strategies, i.e., resource reservation based on threshold and exiting upload upon the length of time slice, were put forward. With these two strategies, nodes with more leftover bandwidth get higher priority to obtain file blocks; be- sides, different file blocks can be delivered to different nodes as soon as POssible. On the basis of LBAPS, a content cloud prototype, P2PStitcher was implemented. Experimental results on PlanetlLab show that the strategies proposed in LBAPS are effective to decrease the average delivery time.