MGPA: a multi-granularity space preallocation algorithm for object-based storage devices

Abstract

Object-based storage systems are promising because they effectively narrow the performance disparity between processors and storage devices. To achieve high performance, object-based storage devices (OSDs) generally preallocate disk space for an object when the desired space is not allocated. However, most existing space allocation algorithms utilise fixed-size preallocation strategies to preserve space for objects, resulting in poor disk space continuity when OSDs concurrently serve multiple objects. In this work, we propose MGPA, an adaptive multi-granularity object space preallocation algorithm to improve the I/O performance of OSDs. MGPA exploits both a user-informed method and an adaptive varied-size method to preallocate disk space. In the simulation-based experimental results, we show that MGPA can significantly improve the object space continuity, which will improve the long-term I/O performance of OSDs.