Effective Resource Scheduling for Collaborative Computing in Edge-Assisted Internet of Things Systems

#Internet #of #things #collaborative #computing #resource #scheduling #game #theory #reinforcement #learning

Along with rapidly evolving communications technologies and data analytics, Internet of Things (IoT) systems interconnect billions of smart devices for performing tasks autonomously, which poses a huge pressure on IoT devices' computing capabilities. Taking advantage of collaborative computing enabled by cloud computing and edge computing technologies, IoT devices can offload computation tasks to idle computing devices and remote servers, thus alleviating their pressure. However, scheduling resources effectively to realize collaborative computing remains a severe challenge due to diverse application objectives, distributed resources, and unpredictable environments. To overcome the above challenges, this thesis aims to design effective resource scheduling for collaborative computing in edge-assisted IoT systems.

To incentivize horizontal collaboration amongst IoT devices, a hierarchical game model is first proposed in smart buildings to obtain maximum utilities for the building management systems and idle computing devices (ICDs), which jointly combines the Stackelberg game and the Cournot game. Under the premise of the subgame perfect Nash equilibrium (SPNE), the BMS can quote the optimal pricing strategy, and ICDs can share the corresponding optimal amount of computing resources. Then, to deal with unpredictability in emergency communication networks, an incomplete information-based two-tier game model is estimated for analyzing the interactions between the emergency management systems (EMS) and ICDs. The Bayesian Nash equilibrium (BNE) is obtained depending on what the EMS and ICDs know. Furthermore, a new computational latency-based pricing scheme is designed from the perspective of the quality-of-experience (QoE) performance, where the computing offloading price varies dynamically with data processing rates. The interactive behaviors between the centralized computing sharing platform (CSP) and ICDs are modeled as the Stackelberg game, seeking out SPNE through the dynamic pricing mechanism, the computation workload selection, and the CPU frequency control. Finally, to meet the increasingly diverse requirements of IoT applications in vertical collaboration between edge systems and IoT devices, a device-specific QoE enhancement resource scheduling scheme is designed, where an online learning approach is proposed on the edge system to schedule communication bandwidth and computational rate simultaneously.