Research On I/O Scheduling Strategy And Data Distribution Scheme In Storage System

In past decades,the performance gap between storage system and CPU is further widened.As the development of multi-core technology,the performance gap will grow.Simultaneously,the Internet and mobile networks increase the amount of data processing and storage.The storage system becomes the bottleneck of computer system.Traditional storage system employs many techniques to improve I/O performance,such as cache,RAID and prefetching mechanism.But it does not fundamentally improve the performance of storage system.Recently,flash memory has the features of low-power consumption,non-volatility,high random access performance and small size;it is well-suited for portable consumer devices.Due to the dramatic price reduction,NAND flash memory has been extending its domain to mass storage systems for PCs or enterprise servers.As the solid-state drive(SSD)does not support in-place update and read-write asymmetry,it cannot completely replace the traditional hard disk drive(HDD).Therefore,the hybrid disk is comprised of HDD and SSD,which uses SSD acts as the cache of HDD.This paper studied the schemes and methods of HDD and SSD,and it aims at improving the performance of storage system.The research topics include:HDD I/O scheduling scheme,NAND flash translation layer(FTL)scheme,SSD-based RAID-6 scheme and data distribution in hybrid disk.The research details of this paper are as follows:First,since the anticipatory scheduling(AS)can't adapt to workload dynamic changes,the Improved Anticipatory Scheduling algorithm is proposed which is based on workload characteristic and service time,short for WPCAS.WPCAS assigns different anticipation period lengths in terms of workload characteristic and decides whether to anticipate the coming request according to service time.Second,most existing flash translation layer(FTL)schemes such as block associative sector translation(BAST)and fully associative sector translation(FAST)will increase garbage collection overhead and degrade the performance of solid-state drive(SSD).A new flash translation layer scheme called Page Write-Related Sector Translation is proposed,short for PWRST.PWRST writes write-related pages into the same data block,which reduces garbage collection overhead and extends the lifetime of SSD.Thrid,SSD-based RAID-6 calculates parity and writes it to disk array when data is updated which decreases the performance of disk array and reduces the life time of SSDs.A partial parity-based RAID-6(short for PRAID-6)is proposed,which uses a delayed parity updating scheme to reduce the number of parity updates.Whenever a page is updated,PRAID-6 generates partial parity data of the stripe and stores it on a special non-volatile memory.The new parity is computed by partial parity and old version of parity when garbage collection mechanism is invoked.Then the new parity of the stripe is flushed to SSD.PRAID-6 reduces the response time of disk array and delays the wear of SSD.At last,by analyzing the workloads from real storage system and considering the advantages and disadvantages between HDD and SSD,the hybrid disk array(HRAID-5)is proposed.HRAID-5 consists of multiple HDDs and one SSD.The data migration scheme and delayed parity updating strategy are applied to HRAID-5,which satisfy capacity,performance and reliability requirements of modern storage system.This paper is focused on the characteristic of HDD and SSD.The experimental results show the feasibility of the schemes and methods.These researches are great significantly in theoretical analysis and practical applications.