Research On I/O Scheduling Strategy For Open Channel SSD

With the rapid development of big data applications,data center I/O latency and throughput have become the focus of attention.Lenovo/IBM’s research report,even with a PCIe-based SSD,when running a typical database system,more than 60% of the time is waiting for external I/O data requests.The existing file system design is mainly based on the sequential access mechanism of the traditional mechanical hard disk,and the data distribution mainly adopts the sequential allocation scheme.The sequential allocation strategy does not fully consider the multi-channel characteristics of the SSD,so that a large amount of data accesses one data channel at the same time,thus seriously wasting data channel resources.Second,the data center typically uses multiple storage nodes for expanded storage capacity,while SSDs have a limited lifetime.Therefore,how to balance the load between multiple storage nodes while maximizing the service life of the entire storage system is one of the urgent problems.To solve the above problems,this paper designs a general SSD model,which not only considers the structure of parallel components in SSD,but also studies the utilization of these components.Then,based on the proposed model,a novel delay-based I/O scheduling strategy is designed.The scheduling strategy not only predicts resource utilization,but also assigns requests to parallel components(e.g,channels)to achieve reduced latency in accessing the SSD.In addition,for the scheme of using multiple storage nodes to expand the storage capacity of the data center,this paper proposes a delay-life-aware I/O scheduling strategy,which not only optimizes I/O performance,but also extends The life of the disk array.Finally,this paper applies the delay-based I/O scheduling strategy and the delay-life-aware I/O scheduling strategy to the trace-driven simulator and evaluates its validity in the Financial and Web Search data sets.Experimental results show that the delay-based I/O scheduling strategy is 16.5% and 14.2% higher than the most advanced scheduling strategy(such as weight-based request scheduling strategy).Second,the latency-life-aware I/O scheduling strategy increases IOPS by more than8% over EGS and improves disk array life.