Abstract:

Large network systems need monitoring to detect flaws and bottlenecks. Many existing monitoring tools only provide overall statistics about global performance metrics. In order to expose the systems internal behavior, we propose a metadata based approach to record the actual execution paths. We have designed proper data structure and information propagation mechanisms to trace the applications internal critical performance-related events. This approach can apply to many network protocols and applications,and it helps capture the sequential and causal relations across domains. Thus it opens a way to diagnose large and complex network systems.

Key words: monitor, tracing, metadata, network