A Voting Scheme for BLOB Replication Preethi Vishwanath
A Voting Scheme for BLOB Replication Preethi Vishwanath Dr. Chris Pollett Dr. Robert Chun Mr. Tuong Truong May 11 th 2007 (10 am – 12 pm)
Outline Brief overview System Design Implementation Phases Bandwidth Usage Byzantine vs. No Replication Strategy Output – Screen shots Comparison with other models
Distance Calculation – Distance between 2 adjacent nodes is 1 Weight = ( distance ) * ( number of accesses )
Our Model U 1 U 9 M 1 U 2 U 8 M 7 1*2 M 2 2*1 4*0 B 1 4*1 2*3 M 3 U 7 1*2 B 1 M 6 4*4 M 4 2*3 U 10 U 6 M 5 U 4 Distance U 5 Accesses
Setup
Operational Flow
Parse XML Store C-XML Parser (expat) <node-details> <mesh-id> 0 </mesh-id> <blob> 0 </blob> <mesh-id> 1 </mesh-id> <blob> 2 </blob> </node-details> 0 0 1 2 Database XML Document Voter Mesh BLOB 1 0 0 1 1 2 … … …
Data Generation BLOB B 1 BLOB access frequency For each BLOB Randomly pick a voter For each BLOB accessed by the voter selected , weight = access * distance Vote r Acces s Distanc e Weight V 1 4*1=4 V 2 2 0 2*0=0 V 3 0 1 0*1=0 V 4 3 2 3*2=6 V 5 6 3 6 * 3 = 18 V 6 1 2 1*2=2
Phase 2 BLOB ∑voters weight B 1 10 B 2 40 B 3 20 B 4 10 Pick Candidate for replication ? for each BLOB { ∑voters weight > α fraction ; }
Basic - Byzantine Agreement For each BLOB being replicated – Initial vote cast ( self machine) – Each node Interested – – – Tabulate votes Toss a coin heads Change vote for next round if more than 5/8 of the nodes agree on a common machine Else change the vote to the majority machine. – tails Change vote for next round if more than 6/8 of the nodes agree on a common machine. Else change the vote to the majority machine. Not interested – Convert vote to majority vote Faulty – Randomly cast a vote. Consensus = 7/8 of the voters agree on the same machine for replication.
Disadvantage M 1 3*2 M 6 B 1 M 2 Tree Model 191 168 2*3 M 5 M 3 3*2 M 4 User 6 = 3 * 2 = 6 User 5 = 2 * 3 = 6 User 4 = 3 * 2 = 6 Since weight of all three users = 6 units Not possible to decide where to replicate 0
Possible Outcomes BLOB replicated to new location No BLOB Replication – No Candidates – No Agreement reached.
Phase 4 - Simulator Extract output information weight = distance * number of accesses BLOB# VOTER_LOC OLD_WEIGHT NEW_WEIGH T
Output Screens Output Byzantine Agreement – Tree Model No Candidates for replication No Byzantine Agreement reached, hence no replication.
No Candidates for replication
No Byzantine Agreement BLOB#, OLD_LOC, NEW_LOC 0, 1, 1 1, 5, 5 2, 7, 7 3, 4, 4 BLOB#, CUM_OLD_BW, CUM_NEW_BW 0, 63 1, 22 2, 20 3, 41
Bandwidth Usage – Ring Topology
Bandwidth Usage – Tree Model
Comparing Byzantine replication with No-replica strategy
Comparison Continuous Broadcast Replication Wait for time t, and then broadcast all the BLOBs requested to all the voters. Static Access Frequency Replication Replicate a copy to the machine which accesses maximum number of times. Dynamic Access Frequency Neighborhood Model – Perform SAF – If the replicated copy is adjacent to the original copy, then replicate on the next highest access.
Continuous Broadcast Replication Bandwidth Comparison Cost-based comparison
Static Access Frequency Replication Bandwidth Comparison/User
Static Access Frequency Replication Cumulative bandwidth
Dynamic Access Frequency Neighborhood Replication Bandwidth Replication
Dynamic Access Frequency Neighborhood Replication Cumulative Bandwidth
References [1] Zune, from Wikipedia the free encyclopedia. http: //en. wikipedia. org/wiki/Zune [2] D. J. Baker, J. Wieselthier and A. Ephremides, “A distributed algorithm for scheduling the activation of links in a self-organizing, mobile, radio network”, Proceedings of IEEE ICC’ 82, 1982. [3] T Hara, N Murakami, S. Nishio, “Replica Allocation for Correlated Data Items in Ad Hoc Sensor Networks”, SIGMOD Record, Vol 33, No. 1, March 2004 [4] T Hara, “Effective Replica Allocation in Ad Hoc Networks for improving Data Accessibility”, Proceedings of IEEE Infocom 2001, pp 1568 -1576. [5] Fundamentals of Database Systems, Fourth Edition, R. Elmasri, S. Navathe, 2003 [6] “Using Expat”, http: //www. xml. com/pub/a/1999/09/expat/index. html [7] S. Jiang, D. He, and J. Rao, “A Prediction-based Link Availability estimation for Mobile Ad Hoc Networks, ” in Proceedings of IEEE Infocom, Anchorage, Alaska, April 2001
Questions ?
- Slides: 27