Hotcold Data Management Strategy For File System On Mobile Devices

NAND flash memory has been used widely in various mobile devices like tablets,smartphones,and MP3 players.Traditional file systems can't take full advantage of flash memory.Flash Friendly File System(F2FS)as a log file system designed specifically for flash memory,perform well on mobile devices.F2 FS uses a heuristics separating hot and cold data,and not aware of the characteristics of data updating on mobile devices,and cannot accurately reflect the temperature of data,which can optimize the data layout of F2 FS,improve read and write performance and reduce GC cost.Experiment shows the file type with the highest update rates on different application are different,and 1 to 2 file types have high updata rate can cover 60% to 80% of the updates.The paper proposes a hotcold data management strategy at the file and page level.At the file level,a file with the file type which has high update rate of application is hot files;at the page level,each page has a temperature-value based on the number of update,which can reflects the temperature of page in real time.And setting a threshold for tempera-ture value to adjust the temperature of page.The threshold of the temperature value can adjust to adapt to the change of temperature of data.At the same time,we propose a new dirty segment selection strategy.We propose a segment-temperature-value for selecting the dirty segment.The temperature value is the sum of page's temperature value which belong to the segment.The segment with lower segment-temperature-value is better for GC.Through testing and analysis,the proposed hotcold data management strategy increase the hot data update rate by 5.23% and the number of hot data pages by 1.8 times compared the semantic-aware.As the main performance gains of hotcold data sepatation,the GC cost was reduced by 20.6%.