A Case Study of Leveraging High-Throughput Distributed Message Queue System for Many-Task Computing on Hadoop

Abstract

Many-Task Computing (MTC) has been a new computing paradigm to address challenging applications that cannot be effectively supported through existing HTC or HPC systems. MTC applications often require a very large number of tasks, relatively short per task execution times, and data-intensive tasks. Each task in MTC applications may require relatively small amount of data processing especially compared to existing Big Data applications typically based on larger data block sizes. However, MTC applications can consist of much larger numbers of tasks where each task communicates through files. Therefore, MTC can be another type of data-intensive workload where a very large number of data processing tasks should be efficiently processed. In order to address MTC type of data processing workflow, we have developed MOHA (Many-task computing On HAdoop) which aims to effectively combine MTC technologies with Hadoop platform. The MOHA framework leverages Hadoop YARN, a popular powerful resource management system, to achieve massive scalability, reliability and effective implementation. In this paper, we report the behaviors of our revised MOHA framework to support real MTC applications on a Hadoop cluster by exploiting high-throughput distributed message queue system (Apache Kafka). Our experimental results show that our approach is effective in handling the real MTC application with the help of robust resource management system provided by Hadoop. In addition, we discuss potential load imbalance problems caused by Kafka and present practical considerations for the MOHA framework to achieve both of high performance task dispatching and good load balancing.