Random Write Performance Improvement For NAND Flash Memory

Nowadays,NAND Flash has been widely used as computer storage to permanently store information.Since the function of the computer applications is becoming increasingly abundant,the demand of space and performance of storage system is getting larger.At the same time,the performance of the computer system and the user experience are greatly depended on the performance of the storage system.Currently,storage systems of computer mainly used NAND Flash to store data.NAND Flash has a good performance on read and sequential write while the random write performance is relatively poor.The primary cause of the bad random write performance is determined by the physical characteristics of NAND Flash,that is NAND Flash can not do in-place update,instead it has to erase the invalid data before reuse the page.According to the research,the frequency of random write is much higher than that of sequential write to some applications.High frequency of random write incurs high frequency of erase operations,thus decreasing the performance of the system.Hence handling the problem of the bad random write performance is critical to improve the performance of computer systems and user experience.This article focused on the random write of NAND Flash and accomplished the following work.First,we analyzed the I/O access features of the workloads.We find that the random write has a large proportion of all the requests.This offers a potential optimization space.This conclusion is the key guide to help us handle the performance bottleneck of NAND Flash.Second,we conducted a survey about the work of this problem in academic circle and analyzed the advantages and disadvantages of this work.The solutions of this problem can be summarized as hybrid storage,that is consider the price,performance,and lifetime of HDD,Flash,NVM(non-volatile memory),and make a combination of these three storages to build a cost-effective storage system.Owing to the immature of technology especially the NVM technology,there is some difficulty to make this scheme practical.It still has the possibility of improvement.Third,according to the research results we propose a new approach to solve the performance problem.We build a hybrid architecture which consists of NAND Flash and NVM.The advantages of this approach are as follows.Firstly,making full use of the in-place update characteristic and the high read/write performance of NVM to service the random writes of NAND Flash.This would greatly reduce the use of NAND Flash space and decrease the frequency of garbage collection.Secondly,making full use of the non-volatility of NVM and safely save the data when power failure occurs.Lastly,we consider the immature of NVM technology,the capacity of NVM can not meet the requirements of fully replacing NAND Flash.We only use a NVM with a small capacity and gain a great performance improvement.Last,we conduct experiments on simulator to verify the effectiveness of the proposed approach.The experimental results show that our scheme can effectively improve the performance of the storage system.