A Self-Organized Task Distribution Framework for Module-Based Event Stream Processing

Abstract

Tackling bottleneck and privacy issues of cloud computing, we attempt to push event stream processing down to devices which are currently empowered to compute and communicate at the edge of the networks. To accomplish that, we propose a self-organized task distribution framework that is composed of multiple brokers collaborating through our module-based event stream processing engine called EdgeCEP. Our system request is event-dependent specified in a brand-new event specification language; still, the event is stored and processed by the relational database. We newly formulate the problem of self-organized task distribution subjective to preferable constraints of computation and communication. The solution for each broker to find individual optimal decision is to apply tabu search with flow-based greedy move regarding pre-ranking flow table. Many experiments are conducted to study and evaluate the performance of the proposed system. The simulation shows that the proposed flow optimization outperforms the naïve algorithm, concretely, 2-times more tasks getting processed and successfully delivered within the same fixed period. The proposed edge-centric method achieves data traffic 7-times less than the cloud-centric approach. The prototype engines have been deployed and evaluated in the real environment.

Keywords

• Stream processing
• complex event processing
• edge computing
• self-organized task distribution
• Internet of Things (IoT)