Meaningful Statistical Analysis of Large Computational Clusters

摘要: Effective monitoring of large computational clusters demands the analysis a vast amount raw data from number machines. The fundamental interactions system are not, however, well-defined, making it difficult to draw meaningful conclusions this data, even if one were able efficiently handle and process it. In paper we show that clusters, because they comprised identical machines, behave in statistically fashion. We therefore can employ normal statistical methods derive information about individual systems their environment detect problems sooner than with traditional mechanisms. discuss design details necessary use these on timely low-impact