Latency-aware Job Dispatching in Fog-cloud Computing Systems

Abstract

As demand grows for low latency services, fog computing is envisioned as a complementary technology to cloud computing, for provisioning distributed computing resources close to end users. The resource limitation at fog servers and the relatively high latency to cloud servers raise the importance of efficient job dispatching strategies. In this thesis, we consider a latency-aware model for fog-cloud computing systems, where jobs arrive in an online fashion, and request services for a period of time with varying resource demands. We take into account separable jobs that can be split into multiple parallel tasks; and the overall latency of a job is determined by the maximum latency of its tasks. The latency sensitivity and priority of each job are characterized by its utility function, and our goal is to maximize the aggregate utility of all jobs. We design polynomial-time online dispatching algorithms, to dynamically dispatch jobs to geographically distributed servers based on utility functions and resource availability, with proven long-term performance guarantees. We evaluate the proposed algorithms by extensive trace-driven simulations, demonstrating their practical effectiveness.