Approximate Parameters Analysis of Closed Fork-Join Queue Model in Object-Based Storage System

摘要

As a new generation of high-performance distributed storage system, object-based storage system is being developed to support high-performance computing environments. In the petabyte-scale object-based storage system, reasonable data distribution and parameters configuration can improve system performance and availability. To make the system performance evaluation work easier, we propose an approximate parameters analysis method to build performance model. We firstly model the whole storage system's architecture based on closed Fork-Join queue model; using our system architecture model, we then deduce an approximately analytical expression with erasure codes and replicas to predict the storage system's mean response time under various workloads simulating the real-world condition. Finally, a large number of comparison experiments validate our approximately analytical expression of system performance, and proved that our analytic method is appropriate to build performance model for object-based storage system.