Download - SOFA communication protocol (EWSN 2014)
1 Challenge the future
SOFA: Communication in Extreme Wireless Sensor Networks
Marco Cattani, M. Zuniga, M. Woehrle, K. Langendoen Embedded Software Group, Delft University of Technology
2 Challenge the future
Motivation
We want to monitor the density of a crowd during an outdoor festival using low-cost wireless devices.. Why?
© Alex Prager
3 Challenge the future
Motivations
• Traditional WSN • Power efficient • Compact hardware
• Low data rate • Slow changes • Few tens of nodes • Sink
We want to monitor the density of a crowd in an open-air festival using low-cost wireless devices
4 Challenge the future
Motivations
• Traditional WSN • Power efficient • Compact hardware
• Low data rate • Slow changes • Few tens of nodes • Sink
We want to monitor the density of a crowd in an open-air festival using low-cost wireless devices
• We are not potatoes!!
5 Challenge the future
Motivations
• Traditional WSN • Power efficient • Compact hardware
• Low data rate • Slow changes • Few tens of nodes • Sink
• We are not potatoes!!
• High data rate • Highly dynamic • Thousands of people • Decentralized
We want to monitor the density of a crowd in an open-air festival using low-cost wireless devices
Extreme Wireless Sensor Networks
6 Challenge the future
Motivations
• Traditional WSN • Power efficient • Compact hardware
• Low data rate • Slow changes • Few tens of nodes • Sink
• We are not potatoes!!
• High data rate • Highly dynamic • Thousands of people • Decentralized
We want to monitor the density of a crowd in an open-air festival using low-cost wireless devices
Extreme Wireless Sensor Networks Communication
7 Challenge the future
Motivations
• Traditional WSN • Power efficient • Compact hardware
• Low data rate • Slow changes • Few tens of nodes • Sink
• We are not potatoes!!
• High data rate • Highly dynamic • Thousands of people • Decentralized
We want to monitor the density of a crowd in an open-air festival using low-cost wireless devices
Extreme Wireless Sensor Networks Communication
8 Challenge the future
Communication challenges
• Bandwidth is trade for energy efficiency
• To reduce bandwidth overhead WSN • exploits neighborhood
information • Synchronize nodes’ wakeups
• Bandwidth is to scarce to be wasted
• We can not rely on neighborhood information
Traditional WSN Extreme WSN
9 Challenge the future
Communication in EWSN
Can we have an efficient rendezvous without neighborhood knowledge?
10 Challenge the future
Communication in EWSN
Init1
34
2
Wakeup period
Yes! But not with unicast and broadcast L n Unicast n Broadcast n Opportunistic anycast
11 Challenge the future
Communication in EWSN
• Efficient rendezvous • Opportunistic anycast
• Collision reduction • Opportunistic rendezvous
• Application layer support
• Contiki OS • LPL and LPP
SOFA (Stop On First Ack) communication protocol Implementation
12 Challenge the future
Efficient rendezvous
More neighbors à Shorter rendezvous n Unicast n Broadcast n Opportunistic anycast
Init1
13 Challenge the future
Efficient rendezvous
More neighbors à Shorter rendezvous n Unicast n Broadcast n Opportunistic anycast
Init12
14 Challenge the future
Efficient rendezvous
More neighbors à Shorter rendezvous n Unicast n Broadcast n Opportunistic anycast
Init1
34
2
15 Challenge the future
Efficient rendezvous
More neighbors à Shorter rendezvous n Unicast n Broadcast n Opportunistic anycast
Init1
34
2
56
16 Challenge the future
Model opportunistic anycast
More neighbors (N) à Shorter rendezvous (R) E(R) = Tw / 1+N (n)
• Nodes’ wake-up period (Tw) • Uniform random variables
• Independent • Identically distributed
• Rendezvous time (R) • First Order statistic
• Beta (1,N)
time
(ms)
neighborhood size50 1000
0
50
100
150
200experimental results
17 Challenge the future
Collision reduction
Transmission Back-Off (TBO) transforms a collision into a successful data exchange
• Listen for incoming beacons instead of CCA
• If a beacon is received, become a receiver
• Less collision among initiators
• Even shorter rendezvous!
InitB B B D A
A D
Rendezvous Data exchange
1
2
Init
TBO
TBO
18 Challenge the future
Information processing
How to cope with the lack of unicast and broadcast?
19 Challenge the future
Information processing
• Select random neighbor • Peer sampling
• Local data exchange • Push-pull • Mass conservation
• Diffuse/aggregate • Max, averages, percentiles
• Repeat until convergence
Gossip
20 Challenge the future
Gossip support
• Select random neighbor to communicate • Neighbor discovery • Random selection
Peer sampling
21 Challenge the future
Gossip support
• Select random neighbor to communicate • Neighbor discovery • Random selection
Peer sampling
Opportunistic peer sampling
• Add random delays to the nodes’ wake-ups
• Use opportunistic anycast to select nodes • No neighbor discovery • Select the most efficient
neighbor (to rendezvous)
22 Challenge the future
Gossip support
• Select random neighbor to communicate • Neighbor discovery +
random selection • Difficult in EWSN
Peer sampling
Opportunistic peer sampling
0 50 1000
200
400
600
800
Node ID
Node
sco
re
ObservedAveragepercentile
23 Challenge the future
Gossip support
2-way data exchange
• Rendezvous once, exchange information twice (2x1) • Improve convergence speed
• Selects quality links • 2-way rendezvous +
3-way handshake
A
D
B B B D
A
24 Challenge the future
Evaluation
020406080
100
card
inal
ity
node positions
L R
Experiment setup
MSP430 CC1101
100
25 Challenge the future
Energy efficiency
duty
cyc
le (%
)
neighborhood size50 5000
0
2
4
6
8
• Settings • Topology: Clique • Message rate: 0.5 • Wakeup period: 1 s • Wakeup time: 10 ms
• Testbed • Size: 5-100 nodes
26 Challenge the future
Energy efficiency
duty
cyc
le (%
)
neighborhood size50 5000
0
2
4
6
8
• Settings • Topology: Clique • Message rate: 0.5 • Wakeup period: 1 s • Wakeup time: 10 ms
• Testbed • Size: 5-100 nodes
• Simulations • Size: 5-450 nodes
The energy consumption of nodes decreases with density
27 Challenge the future
Exchanged messages
glob
al m
essa
ge ra
te (m
sg/s
ec)
neighborhood size50 5000
0
50
100
150
200
• Settings • Topology: Clique • Message rate: 0.5 • Wakeup period: 1 s • Wakeup time: 10 ms
• Testbed • Size: 5-100 nodes
28 Challenge the future
Exchanged messages
• Settings • Topology: Clique • Message rate: 0.5 • Wakeup period: 1 s • Wakeup time: 10 ms
• Testbed • Size: 5-100 nodes
• Simulations • Size: 5-450 nodes
glob
al m
essa
ge ra
te (m
sg/s
ec)
neighborhood size50 5000
0
50
100
150
200
When bandwidth saturates, SOFA continues to exchange messages instead of collapsing
29 Challenge the future
Reliability
deliv
ery
ratio
neighborhood size50 5000
0.90
0.95
1
0.85
• Settings • Topology: Clique • Message rate: 0.5 • Wakeup period: 1 s • Wakeup time: 10 ms
• Testbed • Size: 5-100 nodes
30 Challenge the future
Reliability
• Settings • Topology: Clique • Message rate: 0.5 • Wakeup period: 1 s • Wakeup time: 10 ms
• Testbed • Size: 5-100 nodes
• Simulations • Size: 5-450 nodes
deliv
ery
ratio
neighborhood size50 5000
0.90
0.95
1
0.85
When bandwidth saturates, SOFA continues to reliably exchange messages instead of collapsing
31 Challenge the future
Mobility
• Settings • Topology: Multi-hop • Message rate: 0.5 • Wakeup period: 1 s • Wakeup time: 10 ms • Diameter: ~3 hop
• Simulations • Size: 15-300 nodes • Density: 5-100 nodes
• BonnMotion’s random waypoint • Static (0 m/s) • Walking (1.5 m/s) • Biking (7 m/s)
• Almost identical performance
• Energy efficiency • Exchanged messages • Reliability
Without the need of neighbors’ information, SOFA is resilient to mobility
32 Challenge the future
Does SOFA fulfill our goal?
• Normal conditions • Unicast and broadcast • Routing tree
• Collection
• Aggregation
• Extreme conditions • Opportunistic anycast • Gossip
• Diffusion/Aggregation
• Graph processing
Goal: “We want to monitor the density of a crowd during an outdoor festival using low-cost wireless devices”
33 Challenge the future
Does SOFA fulfill our goal?
• Expected result à
• Legend n 1st percentile n 50th percentile 100th percentile − Data exchange
Demo of SOFA running a gossip protocol to compute in which percentiles nodes’ values are
34 Challenge the future
Does SOFA fulfill our goal?
• Normal conditions • Unicast and broadcast • Routing tree
• Collection
• Aggregation
• Neighbor discovery
• Extreme conditions • Opportunistic anycast • Gossip
• Diffusion/Aggregation
• Graph processing
• Neighborhood size estimation • Poster #8 • Full presentation at IPSN!
Goal: “We want to monitor the density of a crowd during an outdoor festival using low-cost wireless devices”
35 Challenge the future
Conclusions
• Under extreme conditions traditional WSN do not scale
• We proposed SOFA, an opportunistic communication protocol that: • Make an efficient use of bandwidth • Reduce the number of collision • Support gossiping