In this paper, we introduce a new I/O characteristic discovery methodology for performance optimizations on object-based storage systems. Different from traditional methods that select limited access attributes or heavily reply on domain knowledge about applications’ I/O behaviors, our method enables capturing data-access features as many as possible to eliminate human bias. It utilizes a machine-learning based strategy (principal component analysis, PCA) to derive the most important set of features automatically, and groups data objects with a clustering algorithm (DBSCAN) to reveal I/O characteristics discovered. We have evaluated the proposed I/O characteristic discovery solution based on Sheepdog storage system and further implemented a data prefetching mechanism as a sample use case of this approach. Evaluation results confirm that the proposed solution can successfully identify access patterns and achieve efficient data prefetching by improving the buffer cache hit ratio up to 48.24%. The overall performance was improved by up to 42%.

在本文中，我们介绍了一种新的I / O特征发现方法，用于基于对象的存储系统上的性能优化。与选择有限访问属性或大量答复有关应用程序I / O行为的领域知识的传统方法不同，我们的方法可以捕获尽可能多的数据访问功能，以消除人为偏见。它利用基于机器学习的策略（主成分分析，PCA）自动导出最重要的功能集，并使用聚类算法（DBSCAN）对数据对象进行分组以揭示发现的I / O特性。我们已经评估了基于Sheepdog存储系统的I / O特征发现解决方案，并进一步实现了数据预取机制作为该方法的示例用例。评估结果证实，通过将缓冲区高速缓存命中率提高到48.24％，该解决方案可以成功识别访问模式并实现有效的数据预取。总体性能提高了42％。