Multi-queue Round Robin scheduling for enhanced performance in integration platforms

Abstract

Contemporary enterprise environments involve a large amount of information and heterogeneous applications that must exchange data in near real time. Integration platform-as-a-service (iPaaS) solutions support this scenario by executing integration processes composed of workflows of tasks. However, current task scheduling algorithms used in integration platforms, such as First-In, First-Out (FIFO), may lead to poor performance and unfair use of computational resources under high workloads. In this article we propose the Multi-queue Round Robin (MqRR) algorithm, a task scheduling heuristic tailored to the runtime systems of enterprise application integration platforms. MqRR organises tasks into multiple queues and applies a round-robin strategy with preemption to avoid starvation and to distribute the load more evenly among workflows. We evaluated MqRR against the traditional FIFO heuristic using an integration process simulator and three real-world integration workflows, under increasing message arrival rates. Regarding our research questions, the results show that: (RQ1) there is a workload threshold from which FIFO degrades its performance, leading the number of completed messages to approach zero; and (RQ2) MqRR improves task scheduling performance in high workload scenarios, keeping a linear growth of makespan and increasing the number of processed messages. These findings indicate that MqRR is more suitable than FIFO for integration platforms that must handle high message rates in cloud environments.