Hybrid optimization for scheduling budget-constrained energy-aware workflows in the cloud

Cloud computing become a popular environment for deploying compute- and data-intensive scientific applications. However, scheduling workflows in cloud datacenters efficiently to meet all parties’ requirements is still a great challenge. This paper proposes a hybrid optimization approach, GBO+GA, which combines the Gradient-Based Optimizer (GBO) and Genetic Algorithm (GA) for scheduling workflows to minimize the makespan under budget and energy consumption constraints. Firstly, an evolution-aware update strategy switching mechanism is designed to use GBO and GA alternatively in the right time and thus to improve the search diversity. Second, a pattern learning-based auxiliary GBO is adopted to update individuals when the whole population gets stuck into local optima, i.e., replacing the best guiding solution in the original GBO with the defined pattern to make the population escape from local optima and explore more potential areas around. Third, to improve the average fitness of the whole population, some bottom ranked individuals are guided by the global best individual to move towards the optimal solution and speed up convergence. Extensive experiments are conducted on four well-known scientific workflows with different scales, and results demonstrate that GB+GA outperforms its peers, GA, GBO and PSO in the success rate, solution quality and convergence.