AmazingStore: Available, Low-cost Online Storage Service Using - - PowerPoint PPT Presentation

amazingstore available low cost online
SMART_READER_LITE
LIVE PREVIEW

AmazingStore: Available, Low-cost Online Storage Service Using - - PowerPoint PPT Presentation

AmazingStore: Available, Low-cost Online Storage Service Using Cloudlets Ben Y. Zhao Zhi Yang, U. C. Santa Barbara Yuanjian Xing, Song Ding, Feng Xiao Yafei Dai Peking University Motivation Online storage services are getting


slide-1
SLIDE 1

AmazingStore: Available, Low-cost Online Storage Service Using Cloudlets

Zhi Yang, Yuanjian Xing, Song Ding, Feng Xiao Yafei Dai Peking University Ben Y. Zhao

  • U. C. Santa Barbara
slide-2
SLIDE 2

Motivation

 Online storage services are getting increasingly popular

 Amazon’s S3, EMC’s Mozy …  Rely on data centers.

 Challenges

 Threatened by the single point of failure.

 Amazon suffers outages (3 times); Gmail is down (4 times) …  Social networks make downtime harder to hide.

 Incur high hardware, network and cooling costs.

 P2P storage

 Use idle resource of users to avoid costs  Provide low availability because of churn.

slide-3
SLIDE 3

Data Center

Motivation

 Best of both worlds

 Stability of data center  Low cost of P2P

P2P Storage Layer

slide-4
SLIDE 4

AmazingStore Design

 Combine data center and P2P storage system

Master Master Location of replicas Location of replicas

DHT

slide-5
SLIDE 5

AmazingStore Design

 During data center outage

 Degrade to pure P2P storage  Peers closest to servers are assigned as new master nodes.

Master Master Location of replicas Location of replicas

DHT

slide-6
SLIDE 6

AmazingStore Design

 Determine replication degree at the peer layer

 Threshold c derived from hybrid availability model

log(1 ) log(1 ) log(1 )

S P

A c       

Data center availability Peer availability Target Availability

slide-7
SLIDE 7

AmazingStore Design

 Maintain replication level at the peer side

 Probabilistic failure detection  Group-level estimator of replica number

Transient Permanent

Pr(permanent|offline 5 hours) = 0.1 Pr(permanent|offline 25 hours) =0.6 Pr(A and B are transient) = 0.36 Pr(A and B are permanent) = 0.06

Peer A Peer B Peer C

Pr(one is transient, the other is permanent) = 0.58

Possible situations

slide-8
SLIDE 8

AmazingStore: Sharing & Storage

 Upload files important to you

File Storage with guarantee Upload files SLA: Availability guarantee Sharing Files locally with DHT No guarantee

slide-9
SLIDE 9

Preliminary Measurements

 composed of users and data center containing PKU servers.  As of early April. 2010

 Registered users >11,820  Daily peak of online users > 1000  Data objects > 52,055.

 Provides a target of two nines availability

 The data center availability is only 0.932  maintain at least 6 replicas at peer side.

slide-10
SLIDE 10

Availability Improvement

 Overall availability jumps from 93.22% to 99.13%

 Availability gained at peer side is 83.8%

05/25 05/26 05/27 05/28 0.2 0.4 0.6 0.8 1 Date Fraction of available objects Server-side Availability Peer-side Availability System Availability

Power failure P2P layer compensation

slide-11
SLIDE 11

Bandwidth offloading

 90.38% requests were handled by the peer layer.  Average download bandwidth is 2.1MB/sec

06/09 06/11 06/13 06/15 06/17 06/19 06/21 0.2 0.4 0.6 0.8 1 Date Fraction of requests Satisfied on the peer side Satisfed on the server side

Peer-side

  • ffloading
slide-12
SLIDE 12

Discussion

 Edge-cached system cannot work alone.

1000 2000 3000 4000 5000 6000 10 20 30 40 6 Rank Replica count

Fall below the target Only cached replicas

slide-13
SLIDE 13

Discussion

 AmazingStore works well with data recovery

1000 2000 3000 4000 5000 6000 10 20 30 40 6 Rank Replica count

Guarantee enough replicas Cached replicas + Repaired replicas

slide-14
SLIDE 14

Conclusion

 We advocate that data center and peers can complement well.  We describe a deployed prototype called AmazingStore.

slide-15
SLIDE 15

Questions?

Thank you!