Progress on storage systems for disaggregated data centers
Jiwu SHU, Youmin CHEN, Qing WANG, Jing WANG, Junru Li, Xiaojian Liao · Scientia Sinica Informationis · 2023
Exponential growth in data has resulted in unprecedented challenges regarding data storage and management for data centers globally. It is becoming increasingly difficult to meet business needs due to the deficiencies of the traditional server-centric data center architecture in resource utilization, scalability, and performance. Recently, the disaggregated data center architecture has attracted considerable attention from academia and industry. In this architecture, hardware resources are decoupled into different hardware resource pools (such as processor, memory, and storage pools). These resource pools are interconnected through a high-speed network and can be scaled independently as needed; moreover, the pooled hardware resources can be flexibly shared among different applications, resulting in higher utilization. However, the disaggregated data center architecture presents significant differences in memory access mode, storage hierarchy, fault tolerance model, and software overhead, which elicits new challenges in building storage systems on top of such novel architectures. In this work, factors driving disaggregated data centers are analyzed, and their architectural features and advantages are described. Furthermore, the key technologies and representative research work on disaggregated storage systems are summarized. Lastly, future development trends, including memory-level data reliability and heterogeneous computing and networking, are outlined.