Research On Flash Translation Layer Algorithm For Advanced Commands

Flash-based solid state disk in recent years has gradually replaced the traditional hard disk drive due to its desirable properties such as faster read and write speed,lower energy consumption,higher bandwidth,better anti-vibration effect as well as better anti-noise effect.However,it is unable to fulfill the storing,analyzing and processing requirement of big data.Flash-based solid state disk has multiple layers: channel,package,chip,die and plane.These layers provide hardware support of the development of internal and external parallelism of flash.Existing research focuses on improving the external parallelism of flash storage,which is to set up multiple channel controllers in solid state disk to manage the flash channel independently.In addition,some research put forward the corresponding scheduling strategy for internal parallelism of flash,and design the efficient flash translation layer algorithms to develop the plane-level parallelism of flash.Copy back operation is applied in some research to improve the efficiency of garbage collection.However,these studies did not make full use of all levels of flash parallelism which leading to the low resource utilization.By analyzing the internal structure and studying the advanced commands of flash,a flash translation layer algorithm which makes full use of the external and internal parallel of flash present in the architecture of contemporary flash SSDs is proposed,named ADFTL(Advanced Command Flash Translation Layer).The die-level state transfer is applied in ADFTL to develop die-level parallelism by using interleave operation.What's more,we use command generation module to take full advantage of interleave operation and multi-plane operation to make full use of all levels of flash parallelism.In addition,the plane-level block management strategy is presented in ADFTL which selects a special block only used for multi-plane operation.We test the ADFTL in flash SSD simulator called SSDsim,and the results show that ADFTL can greatly reduce the average response time of read and write operations and improve the IOPS.