Quality of Service (QoS)-driven resource provisioning for large-scale graph processing in cloud computing environments: Graph Processing-as-a-Service (GPaaS)

Abstract

Large-scale graph data is being generated every day through applications and services such as social networks, Internet of Things (IoT) and mobile applications. Traditional processing approaches such as MapReduce are inefficient for processing graph datasets. To overcome this limitation, several exclusive graph processing frameworks have been developed since 2010. However, despite broad accessibility of cloud computing paradigm and its useful features namely as elasticity and pay-as-you-go pricing model, most frameworks are designed for high performance computing infrastructure (HPC). There are few graph processing systems that are developed for cloud environments but similar to their other counterparts, they also try to improve the performance by implementing new computation or communication techniques. In this paper, for the first time, we introduce the large-scale graph processing-as-a-service (GPaaS). GPaaS considers service level agreement (SLA) requirements and quality of service (QoS) for provisioning appropriate combination of resources in order to minimize the monetary cost of the operation. It also reduces the execution time compared to other graph processing frameworks such as Giraph up to 10%–15%. We show that our service significantly reduces the monetary cost by more than 40% compared to Giraph or other frameworks such as PowerGraph.