Performance Analyzing System Based on Visualization of Packet Transfers and Method Calls on Object Storage System

Abstract

Object storage systems are widely used for unstructured data such as video data and music data. As these data are increasing its amount, the performance improvement of these object storage systems is increasing their importance. However, understanding their behaviors and finding the bottleneck process in them are difficult because these are distributed systems composed of multiple server computers. In this paper, we evaluate the file uploading performance of the Swift, which is a popular object storage implementation, and show that each uploading takes a long time in some cases. We then propose an analyzing system that visualizes packet transmissions and method calls. The analyzing system identifies the reasons why uploading takes a long time. For evaluating the proposed system, we apply this analyzing method to the Swift and reveal that a process in the Identity Server takes a long time and a hashing function $bcrypt$ is the bottleneck process. Finally, we present a discussion on performance importance based on this analyzing result.