I. Cano*, S. Aiyar, V. Arora, M. Bhattacharyya, A. Chaganti, C. - - PowerPoint PPT Presentation

▶
i cano s aiyar v arora m bhattacharyya a chaganti c cheah
SMART_READER_LITE
LIVE PREVIEW

I. Cano*, S. Aiyar, V. Arora, M. Bhattacharyya, A. Chaganti, C. - - PowerPoint PPT Presentation

Jun 18, 2023 337 likes •612 views

I. Cano*, S. Aiyar, V. Arora, M. Bhattacharyya, A. Chaganti, C. Cheah, B. Chun, K. Gupta, V. Khot and A. Krishnamurthy* Nutanix Inc. *University of Washington NSDI 17 Transparent access to storage Scale up storage by buying more

slide-1
SLIDE 1
  • I. Cano*, S. Aiyar, V. Arora, M. Bhattacharyya, A. Chaganti,
  • C. Cheah, B. Chun, K. Gupta, V. Khot and A. Krishnamurthy*

Nutanix Inc. *University of Washington NSDI ’17

slide-2
SLIDE 2
slide-3
SLIDE 3
  • Transparent access to storage
  • Scale up storage by buying more resources
slide-4
SLIDE 4
  • Automatic replication and recovery
  • Seamless integration of SSDs and HDDs
  • Snapshotting and reclamation of unnecessary data
  • Space-saving transformations
  • …
slide-5
SLIDE 5
slide-6
SLIDE 6

inside

  • Co-design

Reinforcement Learning

slide-7
SLIDE 7

inside

  • Co-design

Reinforcement Learning

slide-8
SLIDE 8

KEY-VALUE STORE I/O SERVICE (foreground) STORAGE MGMT (background)

slide-9
SLIDE 9
  • extents

extent groups

eid egid 33 1984 44 1984 56 1984 Extent Id Map egid physical location 1984 disk1,disk2 … … Extent Group Id Map file 33 44 56 0-8K 8-16K 16-24K 24-32K 32-40K

slide-10
SLIDE 10

KEY-VALUE STORE STORAGE MGMT uses MapReduce I/O SERVICE

slide-11
SLIDE 11
slide-12
SLIDE 12

Node A Node B Node C Node D Curator (A)

Slave

Curator (C)

Slave

Curator (D)

Slave

Curator (B)

Master

Metadata Ring

slide-13
SLIDE 13

Curator (A) Curator (C) Curator (D) Curator (B) 120 -> mtime 120 -> atime (120,mtime,atime)

slide-14
SLIDE 14
slide-15
SLIDE 15
slide-16
SLIDE 16
  • Extensible, flexible and scalable framework
slide-17
SLIDE 17

KEY-VALUE STORE STORAGE MGMT I/O SERVICE Tasks

slide-18
SLIDE 18
slide-19
SLIDE 19

0.2 0.4 0.6 0.8 1 10 20 30 40 50 60 70 80 90 100 Fraction of Clusters SSD Usage (%)

50% of clusters have 75% usage

slide-20
SLIDE 20
  • Extensible, flexible and scalable framework
  • Synchronization between background tasks and foreground I/O
slide-21
SLIDE 21
slide-22
SLIDE 22

AGENT ENVIRONMENT

Action at State st Reward rt st+1 rt+1

slide-23
SLIDE 23
slide-24
SLIDE 24
slide-25
SLIDE 25

5 10 15 20 25 30 35 40

  • ltp
  • ltp skewed
  • ltp varying
  • ltp and vdi

dss

Improvement (%) latency ssd reads

METRIC AVG IMPROVEMENT LATENCY 12 % SSD READS 16 %

slide-26
SLIDE 26

Curator inside

  • Co-designed

Reinforcement Learning