Research On Performance Optimization Of Read/Write Interference Of Flash-based Storage System

With the development of the emerging technologies,such as big data and the Internet of Things,the quantity of information data is growing at an explosive rate.In the meanwhile,data has become more and more complex and diverse,which brings severe pressure and challenges to computer storage systems.Flash-based storage system,with better random access performance than hard disks,has occupied an important position in the high-performance storage systems.However,the read/write interference problem of flash storage is still an important factor affecting system performance under workloads with a mixture of read and write requests.Firstly,to solve the read/write interference problem and significantly improve the read performance,we propose a Hot Read Data Replication scheme for flash storage(short for HotR).HotR utilizes the asymmetric read and write performance characteristics of flash-based SSDs and outsources the popular read data to a surrogate space such as a dedicated spare flash chip or an overprovisioned space within an SSD.By servicing some conflicted read requests on the surrogate flash space,HotR can alleviate,if not entirely eliminate,the contention between the read requests and the ongoing write requests.The evaluation results show that HotR improves the state-of-theart scheme in the system performance and cost efficiency significantly.Moerover,HotR also reduces the tail-latency of the flash-based storage systems.Secondly,we propose a Content-aware Hot Data Replication scheme for Deduplication-based Flash Storage(short for CHDR).CHDR utilizes the content awareness of the deduplication system to identify the hot data with high reference count which is frequently accessed.CHDR outsources the popular data to a surrogate space in advance and then addresses some conflicted read requests to the surrogate apace,thus reducing the read latency caused by the contention between reads and writes and improving the system performance.The evaluation results show that CHDR outperforms HotR in the system performance acceleration.CHDR is more effective for improving the read/write interaction performance of flash-based storage system.The two key technologies for optimizing the read/write interaction performance exploit both the workload characteristics and device characteristics to alleviate the read/write contention of flash storage under workloads with a mixture of read and write requests,thus improving the performance of flash-based storage systems.