Partitioned Real-Time NAND Flash Storage. Katherine Missimer and Rich West

Transcription

1 Partitioned Real-Time NAND Flash Storage Katherine Missimer and Rich West

2 Introduction Eric Risberg AP CircuitsToday 2

3 Introduction Eric Risberg AP CircuitsToday Analytics Vidhya 3 Chesky_W Mapping Ignorance

4 Introduction Eric Risberg AP CircuitsToday Analytics Vidhya The Robotics Institute 4 Chesky_W Mapping Ignorance

5 NAND Flash Memory Non-volatility Shock resistance Low power consumption Fast access time 5

6 NAND Flash Memory Non-volatility No in-place updates Shock resistance Reads & writes operate at different granularity than erasures Low power consumption Fast access time 6

7 SSD Internals 7

8 SSD Internals 8

9 SSD Internals 9

10 SSD Internals 10

11 SSD Internals 11

12 Observations 12

13 PaRT-FTL 13

14 PaRT-FTL 14

15 PaRT-FTL 15

16 PaRT-FTL Buffered pages Page writes for different ways/chips are buffered. 16

17 PaRT-FTL Buffered pages When a block of buffered data is written to each flash chip in Way 1, the write set rotates to Way 2. 17

18 PaRT-FTL Buffered pages 18

19 PaRT-FTL Buffered pages When updating a partially flushed stripe of data, the old pages that are in the buffer are marked as invalid. 19

20 PaRT-FTL Buffered pages 20

21 PaRT-FTL Buffered pages 21

22 PaRT-FTL Buffered pages 22

23 Real-Time Task Model Parameters for a task: # pages in read read request period # pages in write write request period 23

24 Real-Time Task Model Parameters for a task: # pages in read read request period # pages in write write request period A task does not account for the CPU computation time. These tasks exist on the flash translation layer and utilize the NAND bus. 24

25 Real-Time Task Model Parameters for a task: Read capacity: Write capacity: # pages in time to time to read request read a page decode a page 25

26 Real-Time Task Model Parameters for a task: Read capacity: Write capacity: # pages in write request # write time to write a page chips 26

27 Real-Time Task Model 27

28 Real-Time Task Model Encoding task when w > 0: time to encode a page # pages per block # write chips time to write a page 28

29 Real-Time Task Model Encoding task when w > 0: # pages per block write period # pages in write request 29

30 Real-Time Task Model Garbage collection task when w > 0: 30

31 NAND Flash Memory Garbage Collection 1. Write 3 flash pages of Logical Page Numbers (LPN): 0, 1, 2 31

32 NAND Flash Memory Garbage Collection 1. Write 3 flash pages of Logical Page Numbers (LPN): 0, 1, 2 2. Update LPN=0 32

33 NAND Flash Memory Garbage Collection 1. Write 3 flash pages of Logical Page Numbers (LPN): 0, 1, 2 2. Update LPN=0 3. GC triggered to reclaim Block

34 NAND Flash Memory Garbage Collection 1. Write 3 flash pages of Logical Page Numbers (LPN): 0, 1, 2 2. Update LPN=0 3. GC triggered to reclaim Block Copy valid pages in victim block to a free block 34

35 NAND Flash Memory Garbage Collection 1. Write 3 flash pages of Logical Page Numbers (LPN): 0, 1, 2 2. Update LPN=0 3. GC triggered to reclaim Block Copy valid pages in victim block to a free block 5. Erase victim block 35

36 Real-Time Task Model Garbage collection task when w > 0: Over-provisioning 36

37 Real-Time Task Model Garbage collection task when w > 0: Over-provisioning GC victim block 37

38 Real-Time Task Model Garbage collection task when w > 0: time spent copying valid pages and erasing the victim block time before the next block needs to be reclaimed 38

39 Admission Control Read set: Write set: 39

40 Admission Control Read set: time to read a page Write set: minimum read period 40

41 Admission Control Read set: Write set: time to erase a block minimum period in all write, encoding and GC tasks 41

42 Admission Control Simulations # of task sets: 500 # of tasks per task set: 10 Each task makes 1-page read and 3-page write requests per period Periods are calculated based on generated utilizations 42

43 OpenSSD Cosmos Board 43

44 Experimental Results: PaRT-FTL 8 tasks Write requests Read requests 4 tasks make write requests: 12-page writes every 60 msec 4 tasks make read requests: 3-page reads every 15 msec 44

45 Experimental Results: WAO-GC 8 tasks Write requests Read requests 4 tasks make write requests: 12-page writes every 60 msec 4 tasks make read requests: 3-page reads every 15 msec 45

46 Experimental Results 46

47 Conclusion Contributions of PaRT-FTL: An FTL design that takes advantage of internal parallelism in SSDs A real-time task model for read and write requests on multiple flash chips Bounded and low-latency read requests that are not blocked by write requests or garbage collection 47

48 Thank you! 48