Temperature-aware Persistent Data Management For LSM-tree On 3D NAND Flash Memory

Key-value(KV)stores have been widely deployed in both embedded systems and enterprise systems.Today,most KV stores use Log-Structured Merge Tree(LSM-Tree),as LSM-Tree can eliminate random write operations to a storage device and maintain acceptable read performance.As the emerging storage media,three-dimensional(3D)flash memory has become the mainstream technology to replace hard disk drives.Different from hard disk drives and conventional planar flash memories,3D flash memory is vulnerable to temperature.High temperature will cause both charge loss and retention degradation.LSM-Tree was originally designed for hard disk drives.When LSM-Tree is applied to 3D flash memory,LSM-Tree transfers random write operations to the sequential ones.The access to a consecutive physical address in flash memory will cause the temperature issue,which will affect the integrity of the data stored in 3D flash memory.This thesis presents TLSM,a temperature-aware persistent data management scheme for LSM-Tree based KV store on 3D NAND flash memory.TLSM offers both application-level LSM-Tree optimization and firmware-level address management to allocate persistent data to 3D flash.At the application-level,TLSM provides a novel LSM-Tree optimization strategy.For SSTables in LSM-Tree,TLSM separates data and indexes.The separation strategy eliminates unnecessary updates and writes of the LSMTree during compaction,and reduces the amount of data issued from LSM-Tree to 3D flash memory.At the firmware-level,TLSM provides a new address allocation strategy and a new wear-leveling algorithm.TLSM partitions the physical area of the 3D flash memory and assigns logically consecutive addresses to physically dispersed locations.The wear-leveling algorithm allows the data to be evenly distributed throughout the flash memory.The temperature-aware flash physical area allocation and wear-leveling can reallocate the data to physical blocks with relatively low temperature while ensuring uniform data distribution.This cross-layer optimization can effectively handle the temperature issue to ensure the data integrity of the LSM-Tree in the 3D flash memory.We compared the proposed TLSM with Level DB and Wisc Key.Level DB is a very efficient KV store database based on LSM-Tree.Wisc Key optimizes the write performance of Level DB,which can effectively reduce I/O amplification and reduce the amount of data written during compaction.We demonstrated the feasibility of the proposed scheme using a set of standard benchmarks.We compared several performance metrics,including peak and average temperatures,uncorrectable bit errors,system response time,garbage collection overhead,read/write amplification,and the percentage of between read/write operations in compaction among total operations.Experimental results show that TLSM can significantly enhance data integrity and reduce write amplification compared to representative schemes.Compared with Level DB and Wisc Key,TLSM can reduce the peak temperature by 52.99% and 39.61%,and reduce the average temperature by 53.22% and 10.10%,respectively.This cross-layer design can solve the temperature problem of LSM-Tree on 3D flash memory without incurring additional space and time overhead.