220 likes | 336 Views
Data Management in Mobile Ad-Hoc and Sensor Network Databases Le Gruenwald University of Oklahoma and National Science Foundation http://www.cs.ou.edu/~database ggruenwald@ou.edu. Funded by National Science Foundation, NASA and Department of Defense. Two Projects.
E N D
Data Management in Mobile Ad-Hoc and Sensor Network DatabasesLe GruenwaldUniversity of Oklahoma and National Science Foundationhttp://www.cs.ou.edu/~databaseggruenwald@ou.edu Funded by National Science Foundation, NASA and Department of Defense
Two Projects • Data REplication in real-time Ad-hoc Mobile network databases (DREAM) • Window Association Rule Mining to estimate missing data in sensor network databases (WARM)
DREAM: System Architecture • Servers (Large Mobile Host LMH) • Classical workstations with high memory, power and computing capabilities • Contains the complete DBMS • Clients (Small Mobile Host SMH) • Computers with reduced memory, power and computing capabilities • Clients contain the Query Processing Module of the DBMS Server1 Client4 Client5 Client1 Server3 Client2 Server4 Server2 Client3 Client6
DREAM: Research Objective To design and prototype a data replication technique for a Mobile Ad-Hoc Network (MANET) database management system that takes energy restriction, mobility, disconnections and network partitions, and transaction real-time constraints into consideration.
DREAM: Data and Transaction Types Contacted the Norman Fire department and OU Military department for data and transaction model requirements. Data Items Temporal Data Items Read-Only Data Items Persistent Data Items Periodic Update Aperiodic Update Periodic Update Aperiodic Update Transactions Read Transactions Write Transactions Use Current Value Overwrite Current Value MRV MRVP OD Insert/Delete MRV – Most Recent Value MRVP – Most Recent Value in a Partition OD – Outdated Data
DREAM: What and Where to Replicate • Replicates hot data items before cold data items at servers that have high remaining power • Replicates data items that are accessed frequently by firm transactions before those that are accessed frequently by soft transactions • Replicates data items that are accessed frequently by Non-MRV /Non-UCV transactions before those that are accessed frequently by MRV/UCV transactions. • A data item with a higher age relocation ratio is replicated before the one with a lower age relocation ratio. • Reduces replica redundancy between neighboring servers by replacing redundant data items with other unallocated hot data items if the links connecting them is stable.
DREAM: How to Access Replicas • Firm transactions are sent to the nearest server with the least workload • Soft transactions are sent to the highest energy server with the least workload • Read-only data items and Outdated transactions can be executed at any available servers • MRV transactions are executed at the server which has the most recent value of the requested data item • MRVP transactions are executed at the server which has the most recent value of the requested data item within a network partition • Update transactions are forwarded to the primary copy server of the requested data item • Non-UCV transactions are forwarded to any available server if the primary copy server of the requested data item cannot be reached
DREAM: How to Synchronize Replicas Maintains two timestamps: • Primary Update Timestamp – Time when the primary copy server of a data item is updated • Local Update Timestamp – Time when the replica is updated • Server which contains the most recent update timestamp of a data item synchronizes its value with other servers during a certain time interval
DREAM:Performance Evaluation (1) • Compared DREAM with No Replication and Hara’s model • DREAM has more successfully executed transactions and is the most balanced model in terms of energy consumption distribution among servers
WARM: Objective • Develop an efficient data estimation algorithm • to compensate for missing and/or corrupted data ; • to inform intrusion detection component about possible anomalies.
Possible Approaches Possible Approaches Ignore Ask sensor 2 or Estimate the again more sensors missing value Averaging Use existing techniques relations between sensors WARM Approach
WARM: Technical Approach • Mine frequent itemsets from sensor data streams; • Generate association rules according to discovered frequent itemsets; • Use generated association rules to identify sensors related to the sensor MS that has missing/corrupted data; • Develop a weighted function to estimate the missing/corrupted data at MS using the data collected at the sensors related to MS.
Sensor1 Sensor1 Sensor2 Sensor2 MS MS Sensor3 Sensor3 SensorN SensorN WARM: DSARM Framework Apriori Framework DSARM Framework a a b c Find only the rules between pairs of sensors, in which the MS is a consequent, w.r.t. a particular state, satisfying both minSup and minConf Find all valid rules, satisfying both minSup and minConf
WARM • The WARM Approach is an implementation of the DSARM Framework + using an averaging technique to estimate the missing value in the cases association rule mining cannot produce an estimation; • The WARM Approach is a combination of: - the Buffer, the Cube, and the Counter – data structures to store information received by the sensors; - checkBuffer(), update(), and estimateValue() algorithms that use the data in the structures.
WARM: The estimate( )Algorithm Purpose: To estimate a missing sensor reading, using the data in the Counter, the Cube, and the Buffer. • Step 1: Determine eligible states e for MS; • Step 2: Distribute sensors into appropriate eligible StateSets based on their states in the buffer; • Step 3: Determine eligible sensors Si|e-> MS|e; • Step 4: Determine weights for all eligible sensors; • Step 5: Calculate estimated value: missingValue = ( StateValue_e * e) / totalWeight
A collection of 108 sensor nodes deployed on city streets. The sensors collect and report the number of the vehicles detected for a given time interval. sensor nodes server The data is obtained from [AFIDA 03]. B A WARM: Simulation Results (1)
WARM: Simulation Results (2) Evaluation of the Accuracy for different approaches
WARM: Simulation Results (3) Evaluation of WARM TMMAT for different error rates p
WARM: Simulation Results (4) Evaluation of WARM Memory Space Requirements