F-MStorm: Feedback-Based Online Distributed Mobile Stream Processing

摘要

A distributed mobile stream processing system allows mobile devices to process stream data that exceeds any single device's computation capability without the help of infrastructure. It is paramount to have such a system in many critical application scenarios, such as military operations and disaster response, yet an efficient online mobile stream processing system is still missing. In this paper, we make the key observation that the unique characteristics of mobile stream processing call for a feedback-based system design, which is in sharp contrast with the static configuration and scheduling of the current mobile stream processing system, 'MStorm'. At first, we demonstrate the inefficiencies of MStorm through several real-world experiments. Then, we propose F-MStorm, a feedback-based online distributed mobile stream processing system, which adopts the feedback-based approach in the configuration, scheduling and execution levels of system design. We implement F-MStorm on Android phones and evaluate its performance through benchmark applications. We show that it achieves up to 3x lower response time, 10% higher throughput and consumes 23% less communication energy than the state-of-the-art systems.