作者: Olivier Beaumont , Nicolas Bonichon , Philippe Duchon , Hubert Larchevêque
DOI: 10.1007/978-3-540-69355-0_7
关键词: Overlay network 、 Task (computing) 、 Scale (descriptive set theory) 、 Order (ring theory) 、 Cluster (physics) 、 Cluster analysis 、 k-medians clustering 、 Algorithm 、 Context (language use) 、 Mathematics
摘要: In this paper, we consider the clustering of resources on large scale platforms. More precisely, target parallel applications consisting independant tasks, where each task is to be processed a different cluster. context, cluster should enough so as hold and process task, maximal distance between two hosts belonging same small in order minimize latencies intra-cluster communications. This corresponds maximum bin covering with an extra constraint. We describe distributed approximation algorithm that computes resource coordinates i¾? O(log2n) steps O(nlogn) messages, nis overall number hosts. prove provides ratio $\frac{1}{3}$.