Monitoring and performance measuring distributed systems during operation

Abstract

This paper describes an integrated tool for monitoring distributed systems continuously during operation. A hybrid monitoring approach is used. As special hardware support a test and measurement processor (TMP) was designed, which is part of each node in an experimental multicomputer system. Each TMP runs local parts of the monitoring software for its node, while all the TMPs are connected to a central test station via a separate TMP interconnection network. The monitoring system is transparent to users. It permanently observes system behavior, measures system performance and records system information. The immense amount of information is graphically displayed in easy-to-read-charts and graphs in an application-oriented manner. The tools promote an improved understanding of run time behavior and performance measurements to derive qualitative and even quantitative assessments about distributed systems. A prototype of the monitoring facility is operational and currently experiments are being conducted in our distributed system consisting of several MC68000 microcomputers.