hierarchical quorum consensus: a new algorithm for managing replicated data akhil kumar ieee...

21
Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL. 40, NO.9, SEPTEMBER 1991

Upload: abigail-bennett

Post on 04-Jan-2016

213 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data

Akhil KumarIEEE TRANSACTION ON COMPUTERS, VO

L.40, NO.9, SEPTEMBER 1991

Page 2: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

Outline

• Introduction• Quorum Consensus Algorithm• Hierarchical Quorum Consensus• HQC algorithm• Availability Analysis• Tradeoffs between HQC and Related

Algorithm• Conclusion

Page 3: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

Introduction(1/8)

• Motivations of Data Replication1. Fault Tolerant

2. Increasing System Reliability

Page 4: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

Introduction(2/8)1.Providing Fault tolerant capability in distributed system

:One copy of an object

Page 5: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

Introduction(3/8)2.Replication of data for concurrent read/write

The copy is using

:One copy of an object

The copy is using

Page 6: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

Introduction(4/8)

• Two problems occur in distribution system:– RW problem

– WW problem

Read

Write

Read

Write

Write

Write

Page 7: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

Introduction(5/8)

• Two operations of quorum structure in distribution system:– Read operation

• To access all of the copies in a read quorum

• a copy with the highest version number is returned

– Write operation• To write to all of the copies in

a write quorum • assigns each copy the

version number that is one more than the maximum version number encountered in the write quorum.

Read quorum

Write quorum

Page 8: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

Introduction(6/8)

• The solution : intersect property of read/write quorum– RW problem

– WW problem

Read and WriteRead quorum Write quorum

Write and Writewrite quorum Write quorum

Page 9: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

Introduction(7/8)

• This paper generalizes the quorum consensus scheme (QC) – into a multilevel algorithm called hierarchical

quorum consensus (HQC)

– shows that given a collection of n copies of an object, the minimum size of a quorum is n0.63 copies.

• A smaller quorum size results in a lower cost of synchronization.

Page 10: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

Introduction(8/8)

• Our method is based on organizing the copies of an object into – extending the quorum consensus algorithm– Logical node– multilevel hierarchy

Page 11: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

QC Algorithm

• 8 copies let n=8+1 qr+qw > 9 2qw > 9 5 5 4 6

. . . .

• 9 copies let n=9+1 qr+qw > =10 2qw > =10 5 5 4 6

. . . .

Read and WriteRead quorum Write quorum

The quorum intersection conditions:

Read and WriteRead quorum Write quorum

Page 12: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

The concept of HQC

• An example of 2-level

l1=3 l2=3

r1+w1>3 r2+w2>3

2w1>3 2w2>3

2 2 2 2 4 4

1 3 1 3 1 9

1 3 2 2 2 6

r w

best size

Page 13: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

The concept of HQC

Page 14: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

HQC algorithm

For example:l1=3

r1+w1>3

2w1>3

2 2

1 3

Page 15: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

HQC algorithm

Page 16: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

HQC algorithm

=

Page 17: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

HQC algorithm

best size

worst size

Page 18: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

Availability Analysis

HQC

Majority Voting

HQC Majority Voting

HQC

Majority Voting

HQC Majority Voting

Page 19: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

Availability Analysis

HQC

Majority Voting

HQC Majority Voting

HQC

Majority Voting

HQC Majority Voting

Page 20: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

Tradeoffs between HQC and Related Algorithm

HQC is better than others fully.

Page 21: Hierarchical Quorum Consensus: A New Algorithm for Managing Replicated Data Akhil Kumar IEEE TRANSACTION ON COMPUTERS, VOL.40, NO.9, SEPTEMBER 1991

Conclusion

• In this paper, they introduced a new algorithm, also based on voting, and showed that:– It is possible to reduce the size of a quorum

from (n+1)/2 copies (as in majority voting) to n0.63 copies

– The HQC method produces certain intersecting sets of quorums that cannot be produced in a single-level vote assignment