In recent times, cloud computing has become a popular platform for running various applications, with workflows being one of the most common types. However, efficient execution of workflows on cloud resources requires expertise in resource management techniques. Moreover, combining and running workflows from different users can be more cost-effective due to enhanced resource utilization. Therefore, a workflow broker system is necessary to act as an intermediary between users and cloud providers. Although workflow brokers have been extensively studied in recent years, maximizing profits for both brokers and users by utilizing different pricing models has not been adequately addressed. This paper proposes a workflow broker that uses a combination of on-demand and spot instances to minimize workflow execution costs while adhering to deadline constraints. The proposed system classifies resources into multiple classes based on their reliability, with on-demand resources being the most reliable and spot instances classified by their maximum price. The system assigns workflows’ tasks to resource classes based on their criticality, determined by their slack time. Furthermore, to reduce virtual machine provisioning delay, we have utilized container technology to execute workflow tasks. This was achieved by provisioning large virtual machines and executing multiple containerized tasks on each virtual machine. Three different pricing policies are also proposed in this work with the aim of ensuring the broker’s profit, while also offering a reasonable discount to the users. Simulation results indicate that the proposed system not only reduces final costs but also provides higher broker profits and executes more workflows within the deadline compared to previous works.

Keywords