Agent-based Cloud bag-of-tasks execution

摘要

Bag-of-tasks (BoTs) applications are highly parallel, unconnected and unordered tasks. Since BoT executions often require costly investments in computing infrastructures, Clouds offer an economical solution to BoT executions. Cloud BoT executions involve (1) allocating and deallocating heterogeneous resources with possibly different price rates from multiple Cloud providers, (2) distributing BoT execution across multiple, distributed resources, and (3) coordinating self-interested Cloud participants. This paper proposes a novel agent-based Cloud BoT execution tool (CloudAgent) supported by a 4-stage agent-based protocol capable of dynamically coordinating autonomous Cloud participants to concurrently execute BoTs in multiple Clouds in a parallel manner. CloudAgent is endowed with an autonomous agent-based resource provisioning system supported by the contract net protocol to dynamically allocate resources based on hourly cost rates from multiple Cloud providers. In addition, CloudAgent is also equipped with an agent-based resource deallocation system that autonomously and dynamically deallocates resources assigned to BoT executions. Empirical results show that CloudAgent can efficiently handle concurrent BoT executions, bear low BoT execution costs, and effectively scale.