A performance monitoring system for large computing clusters

In this paper we describe the architecture of PerfMC, a performance monitoring system for clusters of workstations; a prototype implementation of the architecture is also presented. PerfMC is driven by an XML configuration file, and uses the Simple Network Management Protocol (SNMP) to collect statistics from each networked equipment. The collected data are maintained on the local disk of the monitoring station in a compact format, and various graphical and statistical analyses can be performed off-line. The monitoring tool embeds an HTTP server which is able to generate various types of graphs from the collected data. Moreover the HTTP server can generate arbitrary XML pages by dynamically applying XSLT stylesheets to an internal XML representation of the cluster's status. The heavy use of XML-based technologies differentiates the proposed approach to traditional monitoring tools.