Font Size: a A A

The Design And Implementation Of Persistent Memory File System NOVA-SSIM With Wear-leveling

Posted on:2020-01-13Degree:MasterType:Thesis
Country:ChinaCandidate:D D HaoFull Text:PDF
GTID:2428330611993645Subject:Electronic Science and Technology
Abstract/Summary:PDF Full Text Request
Novel Non-Volatile Memory,such as 3D XPoint,has the potential to significantly improve the performance of current storage systems and subvert the current storage system architecture.It has the advantages of high bandwidth,large storage capacity and low latency,but at the same time,unbalanced read-write latency and limited lifetime are also the challenges most non-volatile memory needs to face.Most of the research work focuses on improving the performance of persistent memory storage systems.This paper mainly studies how to achieve wear balance of persistent memory media under the premise of low overhead to prolong the lifetime of NVM.It is found that the distribution of write operations on NVM is different with distrinct applications and workloads.Most of them result in imbalance write and lead to hostpot areas in NVM.Besides,the distribution of write varies with the running time,strongly depends on the space management strategy of file system.Based on the above facts and research,this paper proposes a wear balancing oriented persistent memory file system NOVA-SSIM,adopting dynamic wear balancing to achieve balanced distribution of write operations in full persistent memory space,including inode tables,and completely eliminates write imbalance and hot spots.The main contributions of this paper include the following two points:1.A balanced allocation algorithm is proposed to manage the free storage space of non-volatile memory file systemThe main idea is to ensure that all storage spaces are overwritten before reusing a persistent memory page,so as to balance the write access most efficiently.It basically achieves the uniform distribution of write operation requests in persistent memory under a varity of workloads,and slightly improves the system performance.Without metadata,experiment results show that the mean allocation algorithm performs better weal-leveling significantly,the improvement of homogeneity are 67.60%,91.36%,90.82% and 81.03% respectively for Fileserver,Webproxy,Varmail and the Multi-workloads of them,the speed up of performance are 14.09%,0.8% and 1.78%.2.Design and implement a metadata indexing mechanism SSIM,scatter and separate inode mechanism,which is friendly to persistent memory mediaSSIM divides the traditional inode into two parts: the upper part(including inode number)which is updated frequently,and the lower part(other data).The first half is distributed in the whole file system,and is migrated when the update times exceeds certain threshold atomically,rather than locating in fixed position in file system.SSIM completely eliminates hotspots caused by frequent updates of inodes and achieves well-distributed write access in all NVM space.Including metadata,the uniformity of writing access on NVM by Fileserver,Webproxy,Varmai and their mixed loads increased by 96.38%,83.04% and 88.09%,respectively.SSIM has no significant impact on the performance of metadata operation-intensive workloads,like Webproxy and Varmail,but brings 12.4% performance improvement to Fileserver.
Keywords/Search Tags:persistent memory, wear-leveling, memory file system, mean allocation
PDF Full Text Request
Related items