CR: a monitor for distributed systems

Abstract

The increasing popularity of distributed computer systems presents new and important challenges in system management, particularly in a heterogeneous environment. However, there is a lack of tools for managing such systems. This thesis investigates the design and development of a tool for distributed systems. CR is an attempt to create a "control room" where information about behaviour and performance of a distributed system is brought together for inspection, monitoring and review. The CR system architecture consists of five different components, which are data gatherers, data manipulators, display managers, a logging system and an alarm manager running on the same machine. The communication between these components is via two shared memory segments. A data gatherer collects performance data from remote machines, and deposits the data into a shared memory segment. The manipulator retrieves the data, and does further central processing, then deposits the processed data into the second shared memory segment for display. A display manager provides a meaningful, direct view of the performance data. The logging system records performance data into one or more files. It also has a playback tool to display the data from the log files using the display manager. The alarm manager retrieves the performance data, and compares the data with threshold values specified. If the threshold is exceeded, an associated alarm action is taken. CR has a loosely coupled system structure. This feature is good for future development as any changes to one component do not require rebuilding the whole system. At present, CR can monitor systems consisting of different hardware platforms and several different versions of Unix within the distributed system.