~/talk

Scalable Parallel Flash Firmware for Many-core Architectures

Talkat FAST '20 · on Wednesday, 2020-02-26· at 09:00· FAST '20: Grand Ballroom ABCFGH· [embedded-and-hardware]
name:Jie Zhang
links:none on file
links:none on file
links:none on file
links:none on file
slides:none on file

DeepFlash claims a concrete result: a manycore SSD firmware that reaches 1M IOPS and about 4.5 GB/s by splitting firmware into modular “stresses” and running a many-to-many threading pipeline across dozens of lightweight cores. Implementation uses queue-gather → translate → flash-scatter stages, dynamic load balancing of NVMe submission queues, a direct map table to avoid cache lookup bottlenecks, lock-aware translations and offload tricks to cut serialization so flash latency (15% at scale) is no longer the bottleneck.

AI-written summary · about

  1. 0:00Intro
  2. 0:15Executable Summary
  3. 1:34Motivation: Data Explosion
  4. 2:28Motivation: Storage Evolutions
  5. 3:09Baseline NVMe SSD Architecture
  6. 3:41NVMe SSD Firmware
  7. 5:18Scalable Testing - Flash Packages
  8. 5:44Scalable Testing - SSD Performance
  9. 6:36Scalable Testing - Flash Cores
  10. 8:15Modulate firmware as threads
  11. 8:57Modulate firmware into hardware threads
  12. 10:01NVMO thread
  13. 11:05NVMQ thread - dynamic 1/O serialization(DIOS)
  14. 13:11NVMQ thread - consistency issue
  15. 16:45CACHE thread - optimization
  16. 18:52TRANS thread
  17. 20:06Experiment setup
  18. 20:50Overall performance
  19. 21:17SSD controller utilization
  20. 21:40NVMQ thread - load balance
  21. 22:05ILOCK thread - NVMQ parallelism
  22. 23:14Conclusion
~/search
type to search 6,249 conferences, 138,987 speakers and 185,172 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.