A Superblock-based Flash Translation Layer for NAND Flash Memory

Transcription

1 A Superblock-based Flash Translation Layer for NAND Flash Memory Jeong-Uk Kang, HeeseungJo, Jin-Soo Kim, Joonwon Lee Korea Advanced Institute of Science and Technology (KAIST) EMSOFT 06, October 22 25, 2006, Seoul, Korea. Copyright 2006 ACM /06/0010 Yen-Ting Liu

2 Outline Introduction Different kinds of FTL Superblock FTL scheme Performance evaluation Conclusions

3 Introduction Characteristic of NAND flash memory Block and page Block consist of fixed pages Page main data area and spare area Three basic operations read, write and erase

4 Introduction Flash translation layer (FTL) Emulate the functionality of a normal block device with flash memory Hiding the presence of erase operation and erase-before-write Address translation and garbage collection

5 Introduction Address translation Page-mapped and Block-mapped

6 Introduction Classify physical flash memory blocks D-blocks (data blocks) store user data U-blocks (update blocks) handle the erase-before before-write Page management scheme in blocks In-place scheme Logic page number always equal to physical page number Out-of of-place scheme Page can be placed anywhere inside the physical block

7 Introduction Merge operation Full, Partial, or switch merge Full merge Requite copy operation as many as the number of valid page and two erase operations Partial merge Only copy the valid page in D-block D and requite one erase operation Switch merge Requite only one erase operation

8 Different kinds of FTL Replacement block scheme U-blocks and D-blocks D organized by in-place scheme Select a D-block D which has the largest number of U-blocks U as victim Only the partial or the switch merge is performed Log block scheme D-blocks are organized by in-place scheme and U-blocks U by out of place scheme Full merge may happen in order to change from an out-of of-place scheme to an in-place scheme

9 Different kinds of FTL Fully associative sector translation (FAST) A U-block U is shared by all the D-blocksD The full merge may be performed more frequently Superblock FTL scheme A set of adjacent logical blocks that share D-blocks D and U-blocks Allow logical pages within a superblock to be freely located in one of the allocated physical blocks

10 Different kinds of FTL Performance of FTL depend on the efficiency of garbage collection To reduce the numbers of erase operation Increase the utilization of U-blocksU To reduce the number of valid copy To lower the number of full merge operations, while increase chances of partial or switch merge operation

11 Different kinds of FTL Pages are updated in following sequence: P5,P2,P3,P7,P5,P2,P3,P7

12 Superblock FTL scheme Hybrid mapping with three-level mapping table

13 Superblock FTL scheme

14 Performance eveluation Trace-driven simulators for the superblock scheme, the log block scheme and FAST Condition Superblock size is 4 Number of available U-blocks U is 3.1% of D-blocks

15 Performance eveluation

16 Performance eveluation

17 Performance eveluation 0.05% 6.25%

18 Performance eveluation

19 Conclusions The superblock scheme reduces the number of garbage collection invoked, as well as increases chances of switch operation Decrease the garbage collection overhead up to 40% in compared to previous FTL schemes