Performance Analysis and Optimization of Alluxio with OLAP Workloads over Virtual Infrastructure

• Published in: Big Scientific Data Management Vol. 11473; pp. 319 - 330
• Main Authors: Chang, Xu, Liu, Yongbin, Mo, Zhanpeng, Zha, Li
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2019; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Alluxio; Cloud computing; OLAP; Virtualization
• ISBN: 3030280608; 9783030280604
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-28061-1_31

With the popularity of cloud computing, decoupled compute-storage architecture has become a trend. While being able to independently scale compute and storage results in large cost savings and more flexibility, this architecture also increases the latency of data access, reducing the performance. To solve this problem, Alluxio was proposed. Alluxio achieves the goal of reducing data access latency by providing a near-compute cache when Alluxio is deployed with compute nodes. Applications and compute frameworks send requests through Alluxio and are automatically served through the cached copy. Alluxio is used in many production environments and research work, but there is no comprehensive analysis of Alluxio’s acceleration effects. In this paper, we evaluate and analysis the performance of Alluxio with OLAP workloads in different application scenarios. We also summarize the shortcomings of Alluxio and optimize it. Finally, the improved performance results and conclusion are given.