BLog Blocklevel Logblock Management for NAND Flash Memory

BLog: Block-level Log-block Management for NAND Flash Memory Storage Systems LCTES’ 13

Outline • • • Introduction Related Work Motivation Design Evaluation Conclusion

Introduction(1/4) • NAND flash memory is widely adopted in various storage systems – performance – low power consumption – shock resistance. • Challenge – out-of-place updates – endurance

Introduction(2/4) • Flash Translation Layer(FTL)

Introduction(3/4) • Log-block-based FTL schemes have been widely used – Data blocks – Log blocks Data Block Log Block 0 1 1 2 3

Introduction(補充) • Flash Translation Layer(FTL) – Page level mapping – Block level mapping – Hybrid mapping • Data Block (Block level mapping) • Log Block (Page level mapping)

Introduction(4/4) reduce unnecessary write and erase operations, the system performance can be effectively improved

Introduction(補充) • Out Of Band area (OOB) Page datafield OOB This is dedicated for meta information (like information about ECC data, erase counter, etc)

Outline • • • Introduction Related Work Motivation Design Evaluation Conclusion

Related Work(1/3) • Full Merge Data Block Free Block Log Block 0 1 1 2 2 3 Free Block Data Block 0 1 2 3 Free Block

Related Work(2/3) • Partial Merge Data Block Log Block 0 0 1 1 2 3 Free Block Data Block 0 1 2 3

Related Work(3/3) • Switch Merge Data Block Log Block 0 0 1 1 2 2 3 3 Free Block Data Block 0 1 2 3

Outline • • • Introduction Related Work Motivation Design Evaluation Conclusion

Motivation(1/3) • Superblock FTL

Motivation(2/3) • To do the merge operation in Superblock 1, suppose Block 9 is picked up as the victim block Data Block : 0 1 2 Log Block : 9 Full Merge

Motivation(3/3) • From the above example – full merge operations should be avoided – reduce the associativity of log blocks • Present Blog , block-level log-block management scheme

Outline • • • Introduction Related Work Motivation Design Evaluation Conclusion

Design(1/8) • System Architecture

Design(2/8) • Garbage Collection – Situation 1: All spaces of u log blocks allocated to a data block have been used up – Situation 2: the associativity of a log block has reached to l – Situation 3: No free space in log blocks.

Design(3/8) • Situation 1 Log Block 12 Data Free Block 0 12 0 0 1 1 2 2 when the page update request for Page 3

Design(4/8) • Situation 1

Design(5/8) • Situation 2 1. Update request for Page 0 2. Check DLT 3. Select a log block in LMT 4. Garbage Collection

Design(6/8) • Situation 2

Design(7/8) • Situation 3 1. Update request for Page 1 2. Check DLT 3. Select a log block in LMT 4. Garbage Collection

Design(8/8)

Outline • • • Introduction Related Work Motivation Design Evaluation Conclusion

Evaluation(1/3)

Evaluation(2/3)

Evaluation(3/3)

Outline • • • Introduction Related Work Motivation Design Evaluation Conclusion

Conclusion • We proposed a block-level log-block management scheme called Blog – reduce the associativities of log blocks – maximally utilize valid pages and reduce unnecessary erase operations in log blocks.

Thank you for listening