Toward Enhancing Block I/O Performance for Virtualized Hadoop Cluster

摘要

Virtualization offers many advantages by providing server consolidation, isolation, and live migration. In cloud computing environments, virtual machine monitor is an essential part for efficient management of physical resources. Recently, processing big data is important to both most cloud service providers and users. In this paper, the performance of Xen based Hadoop cluster is analyzed, especially related to block device I/O. In addition, the overhead transferring I/O requests are introduced with Xen I/O ring mechanism, and the experimental result of our optimization for Hadoop workloads is presented. The proposed approach reduces CPU utilization by a third during I/O operations and improve throughput of Hadoop applications.