Towards implementing data centric storage for sensor networks

Towards implementing data centric storage for sensor networks Zoltan Padrah doc. dr. Anton Biasizzo

Agenda • Wireless Sensor Networks • Data Centric storage – Distributed Hash Tables • Implementation – Simulator • Performance evaluation • Conclusions

Wireless Sensor Networks Sensor node Gateway node Sensor node User Internet Control centre (web server + data base) • Measure Centralize Offer services

Wireless Sensor Networks Sensor node Gateway node Sensor node Internet User • Centralized architecture • Data is of interest Control centre (web server + data base)

Data Centric Storage Node Internet Sink User Control centre (web server + data base) • All nodes are equal • Any node can accept queries – “What is the temperature in the garden? ”

Data Centric Storage Node Internet Sink User Control centre (web server + data base) • All nodes are equal • Any node can accept queries – “What is the temperature in the garden? ”

Distributed Hash Table • • Mapping (location, time, …) value Stored in a network Hash table: values identified by hashes Node

Distributed Hash Table architecture DHT algorithm Network protocol Network

Implementation steps Prototyping in network level simulator Prototyping in node level simulator Testing on real nodes • Fast development • Easy debugging • Limited accuracy • Not so fast development • Harder to debug • More accurate • Time consuming to debug • Most accurate

Implementation steps Prototyping in network level simulator Prototyping in node level simulator Testing on real nodes • Fast development • Easy debugging • Limited accuracy • Not so fast development • Harder to debug • More accurate • Time consuming to debug • Most accurate

Simulator

Simulator - events • Types of events – External: instructions to the nodes • • Create a node Make a node join a DHT Instruct a node to send query … – Internal: • message passing between nodes • Any process that takes time to complete • Data structure: binary heap At: 0. 001 s Create node 1 At: 0. 002 s Create node 2 At: 0. 010 s node 1 join DHT …

Simulator - nodes • Each node has a unique address – Messages should reach nodes – Search for a given node Address Node structure “address 1” Node 1 “address 2” Node 2 … . . • Data structure: AVL tree – Works well with nodes joining/leaving

Simulator - operation At: 0. 002 s Create node 4 Events At: 0. 001 s Create node 3 Simulator Nodes “addr 2” Node 1 At: 0. 010 s node 1 join DHT “addr 1” Node 2 “addr 3” Node 3

Performance evaluation • Randomly passing messages between nodes 10 9 8 7 Running 6 time 5 4 [s] 3 2 1 0 Execution time vs. number of messages (1000 nodes) 0 100000 200000 300000 400000 Number of messages transmitted 500000

Performance evaluation • Randomly adding and removing nodes 0. 7 Execution time vs. number of nodes 0. 6 (10000 messages) Maximum height of the AVL tree 16 14 0. 5 12 Execution 0. 4 time 0. 3 [s] Maximum 10 height of 8 the AVL tree 6 0. 2 4 0. 1 2 0 0 0 1000 2000 3000 4000 Number of nodes 5000 0 2000 4000 Number of nodes 6000

Performance evaluation • Randomly adding and removing nodes 8 Average looping count for finding one node 7 6 5 Looping 4 count 3 2 1 0 0 1000 2000 3000 Number of nodes 4000 5000

Performance evaluation • Randomly adding and removing nodes 8 Average number of iterations to find a node vs. create/delete events 0. 5 Execution time vs. number of node 0. 45 Create/delete events (1000 nodes) 7 0. 4 6 0. 35 5 0. 3 Average number of 4 iterations in search Execution time 0. 25 [s] 0. 2 3 0. 15 2 0. 1 1 0. 05 0 0 0 10000 20000 30000 40000 Number of node create/delete events 50000 0 10000 20000 30000 40000 50000 Number of delete/create events

Conclusions • Simulator architecture proposed • Efficient data structures used • Solid bases for further development

Thank you for your attention! Questions?