Font Size: a A A

Research On Data Allocation And Management Algorithms For 3D NAND Flash

Posted on:2020-11-30Degree:MasterType:Thesis
Country:ChinaCandidate:Z LuFull Text:PDF
GTID:2428330590958320Subject:Computer system architecture
Abstract/Summary:PDF Full Text Request
In recent years,with the rapid development of information technology,big data intensive applications and data volumes have increased exponentially,which puts higher demands on the capacity,performance and reliability of storage systems.Flash-based storage systems become mainstream storage devices due to their high density,high performance and high reliability.Compared with traditional planar flash memory,3D stacked flash memory uses space stacking technology to organize flash memory cells with higher storage density,which is the development direction of next-generation flash memory.Focusing on the new features of 3D flash,optimizing the data allocation and management algorithms based on 3D flash features can improve the overall performance of the flash memory system,which is a hot spot in the field of flash memory research.Aiming at the problem that the one-shot programming of 3D charge trap flash reduces read parallelism of 3D flash,the one-shot programming aware data allocation policy is proposed,called OSPADA.OSPADA reorders written data to distribute sequential data into different parallel units using the distance aware round-robin strategy.Thus,the multi-level parallelism of flash can be utilized to optimize the flash read performance.Under the 3Dsim simulation platform,OSPADA is evaluated based on the real workload.The experimental results show that OSPADA improves the read performance by up to 22.83% with enhancing 12.18% write performance on average,compared with traditional dynamic data allocation policies.Aiming at the problem that the migration overhead of garbage collection increases due to the increasing number of 3D flash pages,a superblock data management algorithm based on hot and cold data partitioning is proposed,called Dynamic-FTL.Dynamic-FTL firstly divides the 3D flash into hot and cold areas.And then,it performs garbage collection operations in the hot area and thus reduces the number of migration pages of garbage collection by frequently updating the hot data pages.Those reduce the migration overhead of garbage collection and improve the read and write performance while reducing write amplification.Under the 3Dsim simulation platform,Dynamic-FTL is evaluated based on the real workload.The experimental results show that Dynamic-FTL reduces the write amplification by 4.14%,the read latency by 8.5%,and the write latency by 7.14%,compared with the traditional superblock data management strategy.In general,OSPADA and Dynamic-FTL refine the data allocation and management algorithms of 3D flash around the different new features of 3D flash,which improve the read and write performance of 3D flash and reduce write amplification.
Keywords/Search Tags:3D Flash, Data Allocation, Data Management, Garbage Collection, Parallelism
PDF Full Text Request
Related items