Performance monitoring of n-tier client/server systems

Despite the widespread deployment of client/server technology, there seem to be no tools currently available that are adequate for analyzing and tuning the performance of client/server systems. We show that the performance analysis of n-tier client/server systems poses requirements that cannot be satisfied by the currently available monitoring tools for parallel and distributed software. This is mainly due to the fact that the existing tools are limited to the analysis of single, self-contained programs. However, n-tier client/server architectures are built from multiple self-contained programs. Since these programs interact to fulfill a certain task, client/server systems ought to be conceived as a whole in order to fully understand their performance behavior. The segregated, tier-oriented approach to performance analysis fails to address system-wide optimization. We describe the architecture of a performance monitoring tool that has proven appropriate for the analysis of n-tier client/server systems. The tool allows comprehensive instrumentation of multiple programs, enabling it to relate application level events across program boarders. Communication can be monitored at the transport layer, allowing analysis of programs employing arbitrary application protocols. The tool monitors contention for system resources and is capable of tracing certain activities of the operating system kernel, such as CPU scheduling.