Dynamic Scheduling for Seamless Computing

The data generated by Internet of Things (IoT) devices is constantly increasing. The use of cloud computing to process this data is associated with network congestion, has high latency and does not provide context awareness. Fog computing has been proposed as a solution to overcome these problems. The heterogeneous, distributed nature of fog and edge nodes requires a system that facilitates the development and deployment of applications and management of nodes in a cluster. Seamless computing is an extension of fog computing that respects mobility and heterogeneity of nodes. In industry scenarios where cost, energy or network latency optimization is preferred at runtime, static deployment is not sufficient. Production systems require availability, fault tolerance and extensibility, but not at the expense of usability. This research examines the requirements of a system for dynamic rescheduling of software components in a distributed, heterogeneous fog computing cluster at runtime. We propose Dynamic Scheduling for Seamless Computing (DYSCO) as a solution and present a concept implementation based on Kubernetes. We describe the configuration of Kubernetes for DYSCO, extend it with a monitoring tool and enhance the scheduler to enable dynamic components rescheduling. We evaluate the requirements of DYSCO with test cases for an industry-specific scenario in a fog computing cluster. It operates a safety-critical application that must immediately react to machine failures and an application that processes employee data for analytics. We show that DYSCO is able to reschedule software components at runtime, while ensuring technology independence, availability, fault tolerance and usability.