A Study On Buffer Management Algorithms For NAND Flash-based Solid State Drive

With the rapid development of computer architecture technology and chip processing technology in recent decades,the gap between CPU performance and I/O performance continued to expand.At the same time,the rapid development of emerging technologies,such as Cloud Computing,Big Data and Internet of things,also put forward higher requirements on data store.This makes the storage system a performance bottleneck of the computer system.Compared to the traditional hard drives,SSD show many good characteristics,such as high-speed operation,low-power consumption and good reliability,so the SSD has been widely applied in many fields.But SSD also inherits the shortcomings of flash,such as the asymmetry between read operation and write operation,writing-before-erase,the limited erase count,etc.Buffer added in the SSD can cache frequently accessed data effectively,and decrease the number of flash memory write and improve the performance of the SSD.Therefore,the buffer design of SSD has always been a hot research topic.First,the existing page-level buffer algorithms have a higher buffer hit ratio,but will lead to an increase in erase count.The exisiting block-level algorithms are the opposite of the page-level algorithms.So the existing buffer algorithms can't get a certain degree of balance between buffer hit ratios and erase count.Aiming at this problem,this paper proposes a group-based buffer management algorithm(GBBM).GBBM divides the buffer to Page Region and Group Region.Page Region takes the page as the management granularity.Group Region stores the pages that are excluded from the Page Region and aggregates the pages belonging to the same flash block.In this paper,we use SSDSim for simulation experiments.Experimental results show that GBBM performs best when the space of Page Region and Group Region is 3:7.Compared with the page-level algorithms,GBBM makes a slight decline in terms of buffer hit ratio,but makes a significant improvement in terms of erase counts.Compared with the block-level PAB algorithm,GBBM makes a slight improvement in terms of erase counts,but makes a significant improvement in terms of buffer hit ratio.So GBBM gets a certain degree of balance between buffer hit ratio and erase count.Since the 2QW-Clock algorithm does not take into account the relationship between the size of write request and the update frequency,this paper proposes a weight-based buffer management algorithm(WBBM),which inproves the 2QW-Clock algorithm.This paper introduces the Locality Detector in the LAST algorithm and divides the write requests into Small Request and Big Request,which realizes the further distinction among the write requests in the 2QW-Clock algorithm.In this paper,we use SSDSim simulator for simulation experiments.The experimental results show that the performance of WBBM is more pronounced than the 2QW-Clock algorithm for the Financial 1 workload.For other types of workloads,the performance of WBBM and 2QW-Clock is basically the same.