arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1602.01065cs.DCcs.CCcs.NI
收藏

动态图中的分布式在线数据聚集

Distributed Online Data Aggregation in Dynamic Graphs

Quentin Bramas, Toshimitsu Masuzawa, Sébastien Tixeuil

摘要:

我们考虑在动态图中聚集数据的问题,即,将源自图中所有节点的数据聚集到特定的节点,即sink。我们感兴趣的是在不同类型的对手下给出这个问题的下界。在我们的模型中,节点被赋予无限的内存和无限的计算能力。然而,我们假设节点之间的通信是通过成对交互进行的,其中节点可以在决定是否传输数据之前交换控制信息,假设每个节点最多可以传输一次数据。当节点从邻居接收数据时,节点可以将其与其自己的数据聚合。我们考虑三种可能的对手:在线自适应对手,不经意的对手,和随机对手,选择成对的相互作用均匀随机。对于在线自适应和不经意的对手,我们给不可能的结果时,节点没有知识的图,不知道未来。此外,我们给出了几个紧界取决于知识(无论是拓扑相关或时间相关)的节点。对于随机对手,我们表明,收集算法,它总是命令一个节点进行传输,是最佳的,如果节点没有知识。此外,我们提出了一种算法,称为等待贪婪,其中一个节点等待或传输取决于一些参数,这是最佳的,当每个节点知道它的未来成对的相互作用与汇。

英文摘要:

We consider the problem of aggregating data in a dynamic graph, that is, aggregating the data that originates from all nodes in the graph to a specific node, the sink. We are interested in giving lower bounds for this problem, under different kinds of adversaries. In our model, nodes are endowed with unlimited memory and unlimited computational power. Yet, we assume that communications between nodes are carried out with pairwise interactions, where nodes can exchange control information before deciding whether they transmit their data or not, given that each node is allowed to transmit its data at most once. When a node receives a data from a neighbor, the node may aggregate it with its own data. We consider three possible adversaries: the online adaptive adversary, the oblivious adversary , and the randomized adversary that chooses the pairwise interactions uniformly at random. For the online adaptive and the oblivious adversary, we give impossibility results when nodes have no knowledge about the graph and are not aware of the future. Also, we give several tight bounds depending on the knowledge (be it topology related or time related) of the nodes. For the randomized adversary, we show that the Gathering algorithm, which always commands a node to transmit, is optimal if nodes have no knowledge at all. Also, we propose an algorithm called Waiting Greedy, where a node either waits or transmits depending on some parameter, that is optimal when each node knows its future pairwise interactions with the sink.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑