Low-overhead message tracking for distributed messaging

As enterprise applications rely increasingly on commodity messaging middleware, message tracking has become instrumental in testing and run-time monitoring. However, building an effective message tracking system is challenging because of the large scale and high message rate of enterprise-wide applications. To address this challenge, we consider the case of message tracking for distributed messaging middleware. We desire to record the origin, path, and destination of every application message while imposing low overhead with respect to latency, memory and storage. To achieve our goal, we propose a tunable approximation approach based on Bloom filter "histories." Our approach is tunable in the sense that more accurate audit trails may be provided at the expense of storage space, or, conversely, storage overhead is reduced for applications requiring less accurate audit trails. We describe the design of the system and demonstrate its utility by analyzing the performance of a prototype implementation.