Effective Quorum Construction for Consistency Management in Mobile Ad Hoc Networks Takahiro HARA Osaka University, Japan

Data Replication in MANETs Rep. Data replication. ► Preventing deterioration of data availability at the point of network partitioning. Consistency management among replicas!!

Definition: Consistency Consistency of data operations on replicas: All read operations can read the latest version of data items (replicas).

In a MANET, disappearance of nodes and network partitioning frequently occur. ◦ Read-one, Write-all (Completely synchronized) : ◦ Data availability for write operations is very low! Read Write Our previous work [IEEE TMC] - Consistency management (1/2)

Quorum system ◦ Q i : a set of nodes ◦ Every read operation can access at least a latest replica by performing write/read operations to QW ( QR ). = Consistency can be maintained. Write Ex: # of nodes =8, |QW|=5, |QR|=4 Read Our previous work [IEEE TMC] - Consistency management (2/2)

Read Write Quorum system ◦ Q i : a set of nodes ◦ Every read operation can access at least a latest replica by performing write/read operations to QW ( QR ). = Consistency can be maintained. Ex: # of nodes =8, |QW|=5, |QR|=4 Our previous work [IEEE TMC] - Consistency management (2/2) Write/read operations can be performed even if some MHs are disconnected. Data availability can be improved ! [Karumanchi99, Luo03, Hara05]

Goal of our work We address efficient quorum construction and investigate its impact on the system performance. ◦ Investigating the impact of the number of replicas and selection of replica holders (owners) on data availability and traffic.

Effective dynamic quorum construction [Hara’10] System (changeable) parameters: ◦ k j : The number of MHs having the right to replicate D j (replica owners)  Dynamic quorums based on the rule |QW j | + |QR j | > k j ◦ h j : The max. number of MHs in the write quorum to which the write operation is actually performed ( k j ≥ h j ).  Other nodes in the write quorum store the information on the time stamp and the h j nodes (index of the latest replica holder). How to select k j and h j MHs? Access frequency based approach ( k j : static, h j : dynamic)

Performance study (summary) [Hara’10] Message traffic basically depends on k j (larger k j, higher message traffic). h j affects the data traffic for both Write/Read. ◦ Write: larger h j, higher data traffic. ◦ Read: larger h j, lower data traffic. Too small k j shows lower success ratios. k j : Num. of MHs chosen as replica owners h j : Num. of replica owners to which a write is actually performed.

Future directions Dynamic change of replica owners according to the network topology ◦ Checking periodically the number of replica owners in the majority group and changing replica owners.  Protocol: Quorum-based ? Quorum construction considering mobility patterns ◦ Node mobility patterns affect the characteristics of network partitioning, e.g., the number of partitions, partition sizes, and their distribution and stability. ◦ We plan to provide some guideline for effectively constructing quorums.

Selection of k j and h j MHs [AINA’10] How to select k j MHs as replica owners ◦ For each data item D j, MHs with the highest access frequencies are chosen. Criterion: G ij = R ij + W ij ◦ This process is generally conducted once at the config. phase of MANET. How to select h j replica owners to which write is performed. ◦ AF (Access Frequency): aims to reduce data transmission traffic for future Read. Criterion: G ij = R ij + W ij ◦ DIST (Distance): aims to reduce data transmission traffic for the current Write. Criterion: hop counts ( h j closest MHs)

Simulation experiments 100 MHs (M 1,…,M 100 ) exist in an area (X[m]-Y [m], X:Y = 3:4) MHs move according to Random Waypoint. (Max Speed: 2[m/s]) Radio comm. range: 50[m] 500 data items (D 1,…,D 500 ) Read freq.: 0.02[1/s], Write freq.: 0.002[1/s] Read/Write prob. for data items: Zipf Replica allocation method: AF Parameters ◦ Area size: X= [m] ◦ k j : 1, 10, 100, h j : 1, 5, 50 X=area size Y