Starvation Avoidance Task Scheduling Algorithm for Heterogeneous Computing Systems

High-performance and cost-effective Heterogeneous Computing Systems (HCS) are highly dependent on the efficiency of task scheduling, which has been proven to be NP-complete. This makes scheduling problem a vital issue which researchers have attempted optimization in different aspects under the expense of tradeoffs. The problem of scheduling a single Directed Acyclic Graph (DAG) has been extensively researched and studied, however most of the researches in this area did not consider the occurrence of multiple applications represented as multiple DAGs in real case scenarios. This paper aims to fill this gap by providing a fairness-aware, time-efficient, HCS static scheduling algorithm that meets deadlines, reduces makespan, and avoids starvation. The propose algorithm considers the possibility of having to schedule tasks of multiple applications concurrently. The algorithm is called Fair, Deadline-Aware, Time Efficient, Starvation Avoiding (FDATESA) scheduler, and relies on four pillars: (i) mapping the DAGs into information abundant maps (ii) Task priority heuristic calculation (iii) Assigning tasks to processors’ queues (iv) elastic processor time slot calculation for starvation avoidance.