1. On the Effect of Group Mobility to Data Replication in Ad Hoc Networks
Jiun-Long Huang and Ming-Syan Chen
IEEE Transactions On Mobile Computing, May 2006
Presented by Manu Shukla
CS 6204
Fall 2006

2. Agenda The Problem DRAM Algorithm Allocation unit construction phase
VectorCluster
Replica allocation phase
Experiments and Evaluations
Conclusions and Critique

3. Introduction
Mobile Ad Hoc Network (MANET) is a self-organizing, rapidly deployable network of wireless nodes without infrastructure
Mobile nodes of a MANET also function as routers
Disconnection often occurs due to mobility and causes frequent network division
Disconnected partitions decrease data accessibility
Data replication can greatly improve the accessibility for a partitioned network

4. Introduction (2)
DCG and E-DCG are two previously proposed replica allocation schemes in MANET
The two drawbacks of the schemes are:
Generation of large amounts of traffic
Negligence of group mobility

5. Introduction (3)
Authors address the problem by exploring group mobility
Propose Scheme DRAM to allocate replicas by considering group mobility
Underlying group mobility model is assumed to be Reference Point Group Mobility model (RPGM)

6. Description of symbols
Symbols used in formulae and equations

7. Mobility Models
RPGM models team collaboration where mobile nodes collaborate and move as a group
In RPGM, all mobile nodes are divided into several mobility groups
Each node is assigned to virtual reference node and movement of a reference node in a time slot is called global motion vector
The vector from the position of corresponding reference node to mobile node position is random motion vector

8. RPGM Example We have and
where PiN(k) and PiR(k) are positions of the mobile node and reference node in time T(k)

9. System Model m mobile nodes M1, M2,…,Mm and n data items D1,D2,…,Dn
Each data item is updated by its original host periodically with period τi
Each node is equipped with GPS device so its location is always known
Movement of each group follows a waypoint model which breaks movement of mobile node into repeating pause and motion periods

10. DRAM Design
DRAM (Decentralized Replica Allocation with group Mobility) is decentralized algorithm to produce effective replica allocation efficiently
Executed periodically with relocation period r time slots to adapt according to the network connectivity
Two phases in relocation period
Allocation unit construction phase
Replica allocation phase
In allocation unit construction phase, all mobile nodes in network are divided into several disjoint allocation units

11. DRAM Design (2)
In replication allocation phase, the replicas of all data items are allocated according to access frequencies of the data items

12. Allocation Unit Construction Phase
Three mobile nodes states
INITIAL state
ZONE-MASTER and ZONE-MEMBER states
CLUSTER-MASTER and CLUSTER-MEMBER states

13. INITIAL State
Mobile node broadcast info message to all mobile nodes in broadcast zone with a TTL
When a node receives the info message, it forwards it to all nodes that are at TTL or lesser distance from it
Each node maintains a list of its historical locations called a position list to track its pause and motion periods

14. ZONE-MASTER and ZONE-MEMBER states
In ZONE-MASTER and ZONE-MEMBER states
Mobile nodes are classified into two groups by the lowest-id clustering algorithm
Ones with lowest host id are selected as master of their broadcast zone enter ZONE-MASTER state
Other nodes enter ZONE-MEMBER state
Node Mi in ZONE-MEMBER state joins node Mj in ZONE-MASTER state with lowest host id within broadcast zone of Mi

15. ZONE-MASTER and ZONE-MEMBER states (2)
Each node in ZONE-MASTER state then clusters its member nodes
All nodes within a cluster are expected to have similar motion behavior
Master node re-clusters resulting clusters again by considering motion vectors

16. Lemmas
With help of lemmas, we have two heuristics

17. Lemmas (2)
In a mobility group, an actual motion vector is close to the global motion vector if it has
the maximal number of neighbors in angle with maximal difference θ
Maximal number of neighbors in length with maximal difference 2ε
Develop algorithm VectorCluster in accordance with above heuristics

18. VectorCluster VectorCluster consists of two major procedures
ClusterByAngle
ClusterByLength
After executing VectorCluster, each zone master will select one cluster master for each resulting cluster
The selected mobile nodes will enter the CLUSTER-MASTER state, and other nodes will enter CLUSTER-MEMBER

19. VectorCluster (2)
Result of VectorCluster in given example

20. CLUSTER-MASTER and CLUSTER-MEMBER states
Tasks of nodes in this state consist of two steps
Cluster maintenance
Cluster merge

21. Cluster Maintenance
Cluster member sends a status message to its cluster master
Cluster master checks if the moving behaviors similar to one another
It clusters motion behaviors in status messages
Dominating cluster is one with most nodes
It sends reject messages to nodes not in dominating cluster and they return to INITIAL state

22. Cluster Merge
Merging clusters which tend to be connected in the near future improves data accessibility
Two allocation units Ci and Cj can be merged into a new allocation unit if they are cluster wise connected in T(k) and potentially cluster wise connected in T(k+r)

23. Cluster Merge (2)
Here cluster-wise connected and potentially cluster-wise connected are defined as shown
In replica allocation construction, each cluster master will broadcast a merge message containing cluster master id and current and estimate bounding rectangles

24. ClusterMerge Procedure
Cluster Merge can be performed by following process below

25. Replica Allocation Phase
Objective is to
identify data items to be replicated
locations to replicate them for each allocation unit in order to maximize data accessibility
Allocation weight of data item Dj in allocation unit Cx in T(k) is
All data items are allocated in Cx according to their allocation weights in Cx in descendent order
If the candidate set of Dj in Cx is not empty, Dj will be allocated to Mi, where fij is the largest in allocation candidate set of Dj
Allocation process completes if all mobile hosts in Cx is full

26. Procedure ReplicaAllocation
Each master unit then executes ReplicaAllocation procedure

27. Complexity
Complexity of VectorCluster is O(|V|log|V|) where |V| is the number of input vectors
Complexity of ReplicaAllocation is O(m/|c|+n)

28. Integration with other algorithms
Li and Wang proposed RVGM (Reference Velocity Group Mobility)
Yin and Cao proposed scheme RN to balance the tradeoff between data accessibility and query delay
Each mobile node shares only part of its storage with neighbors
A mobile node Mi only cooperates with neighbors which tend to be directly connected to it in future
Easy to integrate these concepts into scheme DRAM

29. Performance Evaluation
Compare DRAM with E-DCG
Use event driven simulator in C++ with SIM Evaluated the performance of DRAM based on several parameters
Assume 120 mobile nodes in a 50mx50m flatland and each node owns 20 data items
Use data accessibility as measure of performance
Accessibility=Number of successful requests/Number of issued requests

30. Performance Evaluation (2)
Use produced network traffic to evaluate cost of schemes
Effect of relocation period below
Shorter relocation period means more executions of relocation schemes making both schemes adapt quickly to relocation behavior of mobile nodes

31. Performance Evaluation (3)
Comparison based on effect of number of Mobility Nodes and number of Mobility Groups
More nodes for same number of mobility groups means more nodes can share their storage by constructing larger allocation units

32. Performance Evaluation (4)
Effect of Number of Replicas per Node
Effect of Update Period
Effect of Precision of Location Information
Effect of Packet Loss Rate

33. Performance Evaluation (5)
Effect of Value of Time-to-Live

34. Conclusions Partitions in MANET frequent problem
Mobility of nodes important consideration for data replication
DRAM algorithm efficient in allocating replicas by considering group mobility
DRAM also produces less network traffic than prior algorithms along with producing higher data accessibility

35. Critique
Introduction to MANET and few examples of disruptive nature of partitioning not adequate
Experiments performed only on simulated data
Lack of real world applications of DRAM and no complexity and performance analysis on real application data a drawback
Number of nodes in simulation relatively small
Consider clustering of moving object techniques similar to ones used in spatial moving objects

36. Q/A?
Thank You!