A Family of Heuristics for Agent-Based Cloud Bag-of-Tasks Scheduling

The scheduling of bag-of-tasks applications (BoTs) in Clouds deal with fixed predefined allocation slots of Cloud resources, e.g., 1-hour time slots, that start being exhausted right after their allocation disregarding whether tasks are being executed or not. In addition, Cloud resources may be allocated for several hours to execute BoTs. However, some resource types (e.g., clusters) may be allocated for only a few hours, while others (e.g., CPU instances) may be allocated for several hours, so BoTs may be executed in such heterogeneous sets of Cloud resources (probably) allocated for a different number of hours. In this paper, a family of 15 scheduling heuristics consisting of two phases: (i) task ordering and (ii) task mapping (based on the remaining allocation times of Cloud resources) is proposed. The heuristics aim to maximize resource utilization while executing BoTs in heterogeneous sets of Cloud resources allocated for different numbers of hours. Cloud resources for executing BoTs are dynamically composed by adopting the contract net protocol. In addition, an agent-based testbed for Cloud BoT scheduling and execution was implemented. Simulation results show that the agents are capable of successfully and efficiently scheduling and executing BoTs in sets of Cloud resources composed from multiple Cloud environments.