Abstract

Fog computing was introduced for the first time by Cisco in 2012. Since then, there has been a great number of studies on fog computing, in which vacant and free-of-charge computing resources in local networks provide low-latency services to end devices.
Even though traditional architecture with scalable and powerful central servers in cloud can accommodate those tasks, it is costly to allocate resources in cloud to execute all those tasks. In addition, it falls short of satisfying Quality of Service (QoS) requirements in terms of waiting time because of long distance communication between servers and user end devices.

In this thesis, we discuss dynamic scheduling problem in fog-cloud collaboration environment for real-time applications when QoS is strict and when an answer is useless if the corresponding application finishes its execution after a pre-defined deadline. By taking into account an admission control procedure to grant only requests whose deadline requirements are feasible with respect to the available resources in the network, we study a proactive scenario using different strategies to calculate schedules and to assign resources, within the admission control procedure to accommodate an incoming request.
Then, we propose our heuristic with four variants corresponding to four different strategies, with the adjustment of a trade-off cost-makespan factor in an utility function. When evaluating performance with some baseline methods in such proactive scenario, the numerical results show that our variants can meet deadline requirements for more applications while exploiting more efficiently the resources in the fog layer and being charged less for using cloud.

Keywords: fog computing, cloud computing, dynamic scheduling, real-time scheduling, task scheduling, workflow applications, DAG, QoS requirements, heterogeneous systems.