Research On Optimization Method Of F2FS File System Semantic-aware Garbage Collection Of Solid State Drives

Random write request suffers from poor storage performance due to small I/O size and discrete storage space.The F2 FS file system is widely used to transform random write to sequential write.However,for flash memory with characteristics such as erase-before-write,the writing unit is a flash memory page and the erasing unit is a flash memory block.Therefore,solid state drives(SSDs)based on flash memory adopts out-of-place updates to write data,and migrate effective data and reclaim invalid data based on garbage collection,increasing additional read/write overhead,causing write amplification problems.Because the F2 FS file system only transfers information to SSDs through a simple block I/O interface,the SSDs cannot obtain the logical block address of the stale data in file system layer and identify data access hotness,so the file system semantic information cannot be effectively used to improve garbage collection efficiency.How to solve the semantic gap between the F2 FS file system and SSDs to improve the efficiency of the garbage collection algorithm is a major research topic that needs to be solved urgently.To solve the problem that SSDs cannot obtain logical block address of the stale data in the F2 FS file system,a stale data-aware garbage collection optimization method for SSDs is proposed.By modifying the Linux I/O stack,logical block address semantic information of the stale data in the F2 FS file system is transferred to the flash memory management layer in SSDs.The flash memory management layer invalidates the logical block address of the stale data,reducing data migration and decreasing write amplification of SSDs.Experimental results show that,compared with traditional garbage collection algorithm,system performance can be improved by up to 118% and write amplification can be reduced by up to 54%.To solve the problem that SSDs cannot recognize the hotness of data access,a hot/cold data-aware garbage collection optimization method for SSDs is proposed.By modifying the Linux I/O stack,the semantic information of data access hotness is transferred from the F2FS file system to the flash memory management layer in SSDs,and then combined with the existing hot/cold identification algorithm,the hot/cold data are stored separately to reduce data migration and decrease write amplification of SSDs.Experimental results show that,compared with garbage collection methods based on hot/cold identification of data logical block address,the system performance can be improved by up to 57%,and the write amplification can be reduced by up to 27%.