| With the advent of the big data era,the growth of data volume is irresistible.The increasing demand for data storage promotes the development of high-density flash mem-ory technology.The high-density flash memory stores more information in a memory cell,reducing storage cost per Gigabyte.But it also comes with a series of problems such as performance degradation and endurance reduction.In order to cover up the defects of high-density flash memory,the solid state drive(SSD)based on high-density flash mem-ory mainly adopts a hybrid architecture.The hybrid SSD with this architecture consists of two regions: low-density flash memory(such as Single-Level Cell,SLC)and high-density flash memory(such as Quad-Level Cell,QLC).Among them,SLC is used as the write cache of the QLC.Although this design hides the poor write performance of high-density flash memory,it does not take the read latency variation into consideration.The previous relevant researches lack the test and analysis on real hybrid SSD.Most researchers solve problems from the controller level and these schemes lack practicality.Therefore,this thesis systematically tests,analyzes the real hybrid SSD and finds key performance characteristics.These features mainly include write features,read features,read-write conflict and capacity features.Based on the above findings,this thesis pro-poses page migration,migration management and request scheduling schemes to make full use of the characteristics of hybrid SSD and optimize the read performance.The main contributions of this thesis are as follows:1.The empirical Study based on hybrid SSD this thesis tests hybrid SSDs from the block I/O layer,file system layer,and application layer,mainly analyzes their read/write characteristics and capacity characteristics,and explores the performance differences between hybrid SSD and non-hybrid SSD.2.Page migration based on read features this thesis proposes a page migration strategy.This design migrates the critical data in the QLC region to the SLC region.In order to reduce the amount of migration,the scheme filters out pages with low latency,and reduces the read latency while ensuring the migration efficiency.3.Migration management based on capacity features this thesis proposes a mi-gration management strategy.This design automatically adjusts the amount of data migration according to the device utilization and avoids migration failure as much as possible.In the data migration phase,the scheme adopts the data block pre-allocation strategy to reduce the fragmentation of the file system and improve the read performance of the storage system.4.Request scheduling based on request features this thesis proposes a request scheduling strategy.The design includes two modules: request merging and re-quest scheduling.In the request scheduling phase,the request is distributed reason-ably according to the page type attribute,read/write attribute,which weakens the interference between requests and further improves the system performance.In the end,a lot of experiments are performed on real hybrid storage devices to verify the effectiveness of the above design.The experimental results show that three methods proposed in this thesis improve the read performance of the storage system to a certain extent. |
