A Study On Buffer Management And Parallel Scheduling Strategy For Solid State Drives

The rapid development of large scale integrated circuit design and manufacturing technology has made the operation speed of the traditional computer which based on von Neumann architecture improve greatly.But the "gap" between the slower memory and the faster CPU has become prominent increasingly,"storage wall problem" has gradually become constraints to enhance the overall performance of the computer system further.SSD(Solid State Drives)consists of solid-state electronic memory chip array and have many advantages such as the fast read and write,light weight,low energy consumption and small size.But the price of SSD is more expensive,read and write speed is different,the capacity is low and the service life is relatively short.SSD have become an important storage device for portable computer systems,desktop computer systems,and server systems.In view of the different access types of different load types,the new buffer management algorithms and the development of internal parallelism which are studied in the flash translation layer of SSD is the key issue to use SSD more fully and effectively,such as the advantage of solid-state hard disk in reading and writing speed,reducing the negative impact of write operations such as high latency,write amplification,etc.In this paper,we propose a write buffer management strategy for SSD to deal with the problem of read and write imbalance of SSD and difficult to make random write request efficiently.The load from network access application has a large number of read requests and random write requests,the write buffer management strategy in this article prioritizes the read request and sets the write cache to hold the write request to adjust the random write request to a sequential write request,thereby improving the solid-state hard disk read and write imbalance and difficult to efficiently deal with random write request,at the same time,we use the heat perception replacement strategy which avoids the situation that the cache hit rate decreases in scheduling process.Experiment based on SSDSim simulator and compare with three typical test sets of Financial,Exchange and CAMWEBDEV.The experimental results show that the write buffer management strategy proposed in this paper reduces the average response time of the access request effectively and reduces the number of erasure of SSD.Secondly,in order to make further use of the parallelism of SSD,this paper proposes a parallel scheduling strategy for SSD.SSD internal have a number of chips,wafers and channels,a reasonable parallel operation can effectively enhance the performance of SSD.This paper presents a strategy that manages the requirements of a write buffer to access the region by dividing the logical space of the solid-state hard disk into multiple zones and creating a dedicated round-robin queue,it use SSD’s good inside parallelism fully.Experiment based on the SSDSim simulator,and Linux system I/O scheduler.The experimental results show that the parallel scheduling strategy for SSD improves the bandwidth of the system access SSD and improves the performance of the system effectively in file I/O and database load.