Abstract
Cloud computing offers an economical and feasible solution for scientific workflow applications requiring large amounts of computational resources and expensive hardware. Supporting Cloud workflow execution involves: (i) allocating and composing a collection of Cloud resources, and (ii) coordinating distributed and self-interested participants. The contributions of this research are: (i) proposing an agent-based approach for supporting workflow execution in one or multiple Clouds, (ii) defining Petri-net based methodologies to design workflows and Cloud resources that sustain concurrent and parallel management of workflows, (iii) implementing an agent-based testbed to simulate distributed workflow execution, and (iv) providing empirical evidence to demonstrate the effectiveness and efficiency of agent-based Cloud workflow execution. The agents are endowed with distributed algorithms, e.g., contract net protocol, to allocate and compose Cloud resources based on workflow requirements. Simulation results demonstrated that: (i) Agents effectively executed (with a 100% success rate) workflows autonomously, even when dealing with concurrent workflow executions, (ii) task parallelization was efficiently achieved in randomly created workflows with different levels of parallelism and ordering constraints, (iii) workflow execution was efficiently achieved since the makespan and number of messages exchanged increased linearly with the number of tasks.