Research On Write Performance Optimization Technologies For Disk Array

With the explosive growth of data,the performance and reliability requirements of storage systems become higher.Disk array technologies have been widely used in highperformance and high-reliability storage systems.However,the write performance of disk array is always a serious problem.RAID-based storage systems have small write performance problem.Logging technologies can improve the small write performance of RAID5.But the logging technologies do not consider the history log when handling write requests,and they are designed based on the characteristics of HDDs.SSDs have many features different from HDDs that are not made full use of.Using SSDs as data disks and HDDs as parity disks to construct hybrid RAID can solve the lifetime problem.But parity blocks on the same disk may result in serious performance bottlenecks.The main research achievements are as follows.To solve the small write performance of RAID5,a history log analysis based write performance optimization for RAID5(HLWO)is proposed.HLWO designs a stripe-based hash table to improve the query efficiency.When choosing reconstruction-write or readmodify-write,HLWO fully considers the history log information to reduce pre-read data blocks number.HLWO not only improves the write performance of RAID5,but also reduces parity re-synchronization time and reconstruction time.When compared with the other two typical logging techniques,the sequential write average response time of HLWO is reduced by 21.86% and 12.91% respectively,and the random write average response time of HLWO is reduced by 33.86% and 16.34% respectively.Due to the logging technologies do not make full use of the characteristics of SSDs,this thesis proposes a new logging technology(RAISL)and a workload characteristics guided access latency for performance optimization in RAIS5(ALO-RAIS5).RAISL makes full use of the invalid data blocks to delay parity blocks update.RAISL only writes the new data to the log disk,and removes the pre-read operations.With the help of RAISL,the workload access characteristics are guaranteed.ALO-RAIS5 exploits the read and write cost characteristic to improve the performance of RAIS5.When write requests access writeonly pages,ALO-RAIS5 uses low latency write.When read requests access read-only pages,ALO-RAIS5 uses low latency read.When requests access interleaved-access pages,ALORAIS5 uses medium latency.The experimental results show that the overall response time of ALO-RAIS5 is reduced by 36.77% and 22.89% respectively when compared with other two RAIS5 performance optimization methods.To improve write performance and reconstruction performance of RAIS5,this thesis proposes an adaptive stripe based reconstruction for RAIS5(ASR-RAIS5).ASR-RAIS5 defines logical stripe and physical stripe.When handling write requests,ASR-RAIS5 uses physical stripe as the basic processing unit.When computing parity blocks,ASR-RAIS5 reduces the number of pre-read operations.When reconstructing,ASR-RAIS5 sikps the failed data blocks that have not been written to improve the speed of reconstruction process.The experimental results show that the write performance of ALO-RAIS5 is improved by 7.92%,and the reconstruction time is reduced by 95.65%,compared with RAIS5.The analysis shows that ASR-RAIS5 improves the reliability of RAIS5.To improve the write performance of hybrid RAID,this thesis proposes a cache based delaying parity blocks update for hybrid RAID(CHyRAID).CHyRAID uses the writethrough mode and writes new data to the cache and hybrid RAID with delaying parity blocks update.CHyRAID exploits the invalid pages to reduce the pre-read operations,and improves the write performance of hybrid RAID.When read requests hit the cache,CHyRAID reads data from the cache.Thus,it also improves the read performance of hybrid RAID.The experimental results show that compared with RAIS4 and a typical hybrid RAID scheme,the overall average response time of CHyRAID is reduced by 43.77% and 38.22% respectively.