maekawa: quorum size research jeremy miller kent state university november 28 th, 2011 1

26
Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th , 2011 1

Upload: robert-strickland

Post on 18-Jan-2018

219 views

Category:

Documents


0 download

DESCRIPTION

Outline Overview of Maekawa Quorum Creation Experimental Setup Results Future Work Conclusions 3

TRANSCRIPT

Page 1: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Maekawa: Quorum Size Research

Jeremy MillerKent State UniversityNovember 28th, 2011

1

Page 2: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Outline•Overview of Maekawa•Quorum Creation•Experimental Setup•Results•Future Work•Conclusions

2

Page 3: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Outline•Overview of Maekawa•Quorum Creation•Experimental Setup•Results•Future Work•Conclusions

3

Page 4: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Overview of Maekawa•Permission based DMX solution•Each process has a quorum•Each process only has one permission to give•Process is granted access to the critical section

if it receives permission from its entire quorum•6 Message types:

▫Request, Permission, Release Message Complexity: 3√N

▫Failure, Inquire, Yield (Deadlock Messages) Message Complexity: 6√N

4

Page 5: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Outline•Overview of Maekawa•Quorum Creation•Experimental Setup•Results•Future Work•Conclusions

5

Page 6: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Billiards Algorithm

1st Path2nd Path 3rd Path

1st Path 2nd Path3rd Path

Where to start

Where to start

Quorums:Quorum[0]= { }Quorum[1]= { }Quorum[2]= { }Quorum[3]= { }

6

Page 7: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Billiards Algorithm

1st Path2nd Path 3rd Path

1st Path 2nd Path3rd Path

Where to start

Where to start

Quorums:Quorum[0]= {0, 1, 2}Quorum[1]= { }Quorum[2]= { }Quorum[3]= { }

7

Page 8: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Billiards Algorithm

1st Path2nd Path 3rd Path

1st Path 2nd Path3rd Path

Where to start

Where to start

Quorums:Quorum[0]= {0, 1, 2}Quorum[1]= {1, 2, 3}Quorum[2]= { }Quorum[3]= { }

8

Page 9: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Billiards Algorithm

1st Path2nd Path 3rd Path

1st Path 2nd Path3rd Path

Where to start

Where to start

Quorums:Quorum[0]= {0, 1, 2}Quorum[1]= {1, 2, 3}Quorum[2]= {0, 2, 3}Quorum[3]= { }

9

Page 10: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Billiards Algorithm

1st Path2nd Path 3rd Path

1st Path 2nd Path3rd Path

Where to start

Where to start

Quorums:Quorum[0]= {0, 1, 2}Quorum[1]= {1, 2, 3}Quorum[2]= {0, 2, 3}Quorum[3]= {0, 1, 3}

10

Page 11: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Non-Billiards Algorithm•Make a matrix

counting upwards

•Find the process

•Draw two lines, add all of the processes hit to the quorum.

0 1

2 3

Quorums:Quorum[0]= { }Quorum[1]= { }Quorum[2]= { }Quorum[3]= { }

11

Page 12: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Non-Billiards Algorithm•Make a matrix

counting upwards

•Find the process

•Draw two lines, add all of the processes hit to the quorum.

0 1

2 3

Quorums:Quorum[0]= {0, 1, 2}Quorum[1]= { }Quorum[2]= { }Quorum[3]= { }

12

Page 13: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Non-Billiards Algorithm•Make a matrix

counting upwards

•Find the process

•Draw two lines, add all of the processes hit to the quorum.

0 1

2 3

Quorums:Quorum[0]= {0, 1, 2}Quorum[1]= {0, 1, 3}Quorum[2]= { }Quorum[3]= { }

13

Page 14: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Non-Billiards Algorithm•Make a matrix

counting upwards

•Find the process

•Draw two lines, add all of the processes hit to the quorum.

0 1

2 3

Quorums:Quorum[0]= {0, 1, 2}Quorum[1]= {0, 1, 3}Quorum[2]= {0, 2, 3}Quorum[3]= { }

14

Page 15: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Non-Billiards Algorithm•Make a matrix

counting upwards

•Find the process

•Draw two lines, add all of the processes hit to the quorum.

0 1

2 3

Quorums:Quorum[0]= {0, 1, 2}Quorum[1]= {0, 1, 3}Quorum[2]= {0, 2, 3}Quorum[3]= {1, 2, 3}

15

Page 16: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Outline•Overview of Maekawa•Quorum Creation•Experimental Setup•Results•Future Work•Conclusions

16

Page 17: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Billiards vs. Non-Billiards•Testing 2 parameters: Message Complexity and

Correctness in critical section entry.•Hypothesis: Billiards will have smaller message

complexity but a worse correctness in CS entry.•Methods:

▫Set up 2 fixed quorums for each type. Run increasingly large sample sizes of CS entry. Run each test 10 times and average the results.

▫Billiards: 24 Processes, 7 Quorum Size.▫Non-Billiards: 25 Processes, 9 Quorum Size

17

Page 18: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Outline•Overview of Maekawa•Quorum Creation•Experimental Setup•Results•Future Work•Conclusions

18

Page 19: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Results

19

Page 20: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Results (Cont.)

20

Page 21: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Results (Cont.)•On all tests done the correctness of CS

entry was 100% for both Billiards and Non-Billiards

•Not once was there incorrect order in entry.

•Expected a wrong entry in the first 1-3 entries every once in a while.

21

Page 22: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Outline•Overview of Maekawa•Quorum Creation•Experimental Setup•Results•Future Work•Conclusions

22

Page 23: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Future Work•Test different sizes of quorums, not only 2

fixed sizes.

•Force the algorithms into out-of-order situations initially and see if there is any difference in correctness between the two algorithms.

23

Page 24: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Outline•Overview of Maekawa•Quorum Creation•Experimental Setup•Results•Future Work•Conclusions

24

Page 25: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

Conclusions•Billiards algorithm is the all-around best

algorithm for quorum creation

•Both algorithms level off in messages sent around 500 critical section entries

•No advantage to use anything other than billiards quorum creation

25

Page 26: Maekawa: Quorum Size Research Jeremy Miller Kent State University November 28 th, 2011 1

References•Agrawal, D. "Billiard Quorums on the

Grid." Information Processing Letters 64.1 (1997): 9-16. Print.

•Code Defense Link: Code Defense•Thank you.•Questions?

26