Multi-dimensional storage QoS guarantees for an object-based storage system

Abstract

The Object-based storage is an emerging storage architecture that could easily fulfill multi-dimensional storage QoS requests. This paper focuses on providing QoS guarantees under Object storage infrastructure along the three most prevalent dimensions: capacity, bandwidth and latency through storage resource allocation and IO commands scheduling. Firstly we propose an algorithm on storage resource mapping derived from Toyoda algorithm, which achieves efficient resource utilization through consideration of the USDs' serving ability. Secondly we propose an object commands scheduling mechanism and develop a prototype system based on the Lustre filesystem. Through adding timestamp to each object command and scheduling the command queue by final finish time, the system can efficiently fulfill the demands on latency from the front applications. © Springer-Verlag Berlin Heidelberg 2006.