Design And Implementation Of Cache Subsystem In Distributed Block Storage System

Traditional storage technologies,such as NAS and SAN,have low performance and pool scalability due to the centralized architecture.It is hard for them to meet the large-scale storage requirements in cloud computing.In recent years,standard server based and software-defined distributed block storage systems have achieved great success.They are able to easily scale out and have many advantages over traditional storage technologies,thus becoming the mainstream in industry and the hotspot of research.Meanwhile,new storage devices are revolutionizing a new generation of storage systems.SSDs eliminate the mechanical movement in HDDs and provide significant performance boost in random I/O,making them an effective way of performance optimization.Based on the study of large-scale distributed storage systems,This paper proposes a SSD and NVRAM based design of distributed cache system for block storage system BoSS.The main work and innovation of this paper is as follows:(1)A SSD-based client-side distributed cache system is designed and implemented.It not only improves I/O performance but also reduce communication cost and CPU overhead.Besides,NVRAM is combined to solve the erasure life problem of SSDs.(2)The reliability of the cache system is ensured by using multi-replica technology and strong consistency I/O model.(3)A two-level distributed metadata management method is proposed for the large-scale metadata in the cache system,which provides high performance and reliability.(4)According to different characteristics of data access,a well-directed data heat management strategy is designed to avoid the interference between read and write,and,cache pollution by sequential access.(5)The improved ARC algorithm is designed and implemented,and the efficient cache selecting and replacing strategy is implemented by fine-grained data heat tracing.(6)The cache system is optimized by data deduplication and cache migration flow control for logical volume snapshots and virtual machine migration.The experimental results show that the cache system designed and implemented in this paper can effectively reduce the I/O delay and client-side network related overhead in distributed block storage system BoSS,achieving effective performance optimization in cloud computing scenario.