Scheduling for Heterogeneous Computing Platforms Using a Genetic Algorithm

In heterogeneous computing platform interconnected by high-speed networks, the task scheduling problem has been extensively studied. Such systems promise to quickly handle computationally intensive applications with different computing needs. The HEFT algorithm is proposed to solve the task scheduling problem of such a system. The HEFT algorithm is simple but its efficiency needs to be improved. The basic idea of our method is to take advantage of genetic algorithms and HEFT algorithms while avoiding their disadvantages. The algorithm uses the HEFT algorithm to assign priorities to each subtask, while using a genetic algorithm to search for a task-to-processor mapping solution. The GHEFT method also designs crossover, mutation, and fitness functions suitable for directed acyclic graph (DAG) scheduling. Experimental results show that the GHEFT algorithm is superior to a non-evolutionary heuristic algorithm and a random search method in terms of scheduling quality.