Research On Wear-aware Persistent Memory File System Space Management Optimizatio

Persistent memory file systems utilize persistent memory(PM)features to optimize file data organization and data consistency mechanisms,achieving significant improvements in system performance.However,existing persistent memory file systems often overlook issues such as the low write endurance of persistent memory.Inappropriate space management strategies can easily wear out persistent storage,severely compromising data reliability.Meanwhile,existing space management techniques that account for wear degrees still suffer from low wear-leveling accuracy and high overhead.This thesis proposes a Wear-Leveling-Aware Buddy-Like Allocator(WBAlloc)to improve wear-leveling accuracy and system performance.WBAlloc consists of two key technologies,Multi-Level Allocation Mechanism(MLAM)and Wear-Aware Reclamation Mechanism(WARM).MLAM uses a multi-level table based on granularity range grouping to manage persistent memory free space without redundant splitting caused by insertions of free blocks,and accelerates the location of available lists based on the availability bitmap,as well as an adaptive insertion mechanism based on the average wear degree to keep the free blocks relatively ordered with low overhead.The WARM constructs a reclaim chain to manage the released free blocks and ensure the balance of wear within the free blocks based on the intra-block page wear difference and utilizes a free block locator to accelerate the block remlamation and merging process.When there are few free blocks in MLAM,WARM migrates the less-worn free blocks to MLAM.Finally,this thesis implements a prototype for WBAlloc by modifying the kernel source code of a typical persistent memory file system,NOVA,and integrating it into the Linux kernel.Compared to NOVA,DWARM,and WMAlloc(two state-of-the-arts wear-aware allocators for persistent memory file systems),experimental results show that WBAlloc improves the average system performance by 26.23%,80.46%,and15.61%,respectively,and the average persistent memory lifetime by 338.88%,159.28%,and 29.45%,respectively.