Scheduling Bag-of-Tasks Applications with Budget Constraints on Hybrid Clouds

In this paper, we study budget-constrained Bag-of-Tasks (BoT) scheduling problem on hybrid clouds to minimize makespan. Each BoT application consists of multiple tasks, each of which has different runtimes corresponding to different Virtual Machine (VM) types provided by cloud providers. This problem is formulated as an integer programming, which is generally NP-Hard. To solve the formulated problem, we propose four effective heuristics, in which task sequences are regarded as scheduling solutions. We develop a Sequential-Runtime-based Arrangement (SRA) method to generate the initial solution of each proposed heuristic. In SRA, tasks are arranged by a metric called sequential runtime instead of their runtimes corresponding to different VM types. Four well-known neighbourhood search methods are employed and iterated to improve solutions, respectively. Solutions' objectives are calculated by a Greedy Dispatching (GD) mechanism, which is able to schedule tasks effectively for minimizing makespan without violating the budget constraint. By exhaustive experiments, high effectiveness and efficiencies of proposed heuristics are verified and two observations are found.