Research Of NVM-based File System Mobi-PMFS Based On Mobile Application Characteristics

Recently,the development and maturity of technologies such as mobile Internet,edge computing,Internet of things have born many new mobile devices and mobile applications,such as wearable devices with health monitoring function,mobile e-commerce,network mobile game,short video etc.These emerging mobile applications have higher data access and computing requirements,and show different data access patterns and data distribution characteristics.Small file random read write operations account for a large part of data access in these new mobile applications,This large number of fine-grained storage requirements leads to the fact that the mobile device storage system based on block devices is not enough to meet the storage requirements of these applications.Research shows that the performance bottleneck of mobile devices has changed from the network and processor to the data access.The emerging non-volatile memory(NVM)has many advantages,such as byte-addressability,high memory density,near-DRAM speed etc.These new features coincide with the need for fine-grained access of these mobile applications,however,existing persistent memory file systems based on non-volatile memory do not well adapt to the application characteristics of mobile devices,resulting in the advantages of these persistent memory file systems cannot be maximized in mobile devices.Based on the storage performance challenges in mobile devices and new technology,this thesis applies the new NVM to mobile devices and proposes a new storage architecture for mobile devices.The ARM architecture adapter is designed and implemented to transfer efficient persistent memory file system PMFS to mobile system to improve the storage performance of mobile devices.Considering the current persistent memory file system is not well adapted to the mobile application access characteristics and cannot make good use of its performance advantages.This thesis analyzes the access trace of mobile applications and then establishes a mobile application feature model,through which a mobile application feature aware three-lists space management allocation strategy is proposed and deployed to Mobi-PMFS.The three-lists are the traditional 4KB granular allocation list,the 2KB granular allocation list to meet the fine-grained access requirements of mobile applications,and the 1MB granular allocation list to meet the new mobile applications allocation requests for contiguous large physical pages.Secondly,considering the data distribution model of mobile applications and the write limited endurance of NVM,this thesis implements the hot cold zone conversion list switching algorithm in the three-lists allocation strategy to realize the wear-leveling on NVM.In addition,considering that different user preferences may lead to the decrease of Mobi-PMFS space allocation efficiency,a list reviser is proposed to ensure the effectiveness of the three-lists strategy.Finally,a simple energy consumption model is established for the traditional storage architecture of mobile devices and the new storage architecture based on NVM proposed in this thesis.The data shows that the new storage architecture can reduce the storage energy consumption of mobile devices by more than 80%.The experiment shows that the read and write throughput of Mobi-PMFS is better than traditional mobile storage system by4 x,and the execution time of typical mobile application is reduced by 95%.The wear-leveling strategy of three-lists can effectively prolong the lifetime of NVM of mobile devices.