310 likes | 430 Views
INFOCOM 2004 – Hong Kong. Modeling the Performance of Wireless Sensor Networks. Carla Fabiana Chiasserini Michele Garetto Telecommunication Networks Group Politecnico di Torino, Italy. Outline. Network Scenario Our contribution Modelling approach Sensor model Network model
E N D
INFOCOM 2004 – Hong Kong Modeling the Performance of Wireless Sensor Networks Carla Fabiana Chiasserini Michele Garetto Telecommunication Networks Group Politecnico di Torino, Italy
Outline • Network Scenario • Our contribution • Modelling approach • Sensor model • Network model • Interference model • Numerical results • Conclusions and future work
Network scenario • Large number of self organizing, unattended micro-sensors • Short radio range multi-hop wireless communications towards a common gateway • Energy-limited (battery operated) • Sleep/active dynamics • Energy efficiency is the crucial design criterion
Our contribution • Analytical model to predict the performance of a wireless sensor network • Responsiveness (data transfer delay) • Energy consumption • Network capacity • Our model combines together • Sleep / active sensor dynamics • Channel contention and interference • Traffic routing • An analytical approch to understand fundamental trade-offs and evaluate different design solutions
Modelling approach • Sensed information is organized into data units of fixed length • Time is slotted • slot = time needed to transfer a data unit between two nodes (including channel access overhead) • discrete time model • Data units are generated by each sensor at a given rate (during active period) • Data units can be buffered at intermediate nodes (infinite buffers)
Reference scenario sensor gateway N = 400 sensors randomly placed (uniformly) in the disk of unit radius
System solution Model decomposition SENSOR MODEL NETWORK MODEL INTERFERENCE MODEL Iterate with a Fixed Point Solution
Sensor model: assumptions ACTIVE Generation of new data units Transmission of data units Reception of data units R S SLEEP N Transmission of data units ~ geom(p) ~ geom(q) R S R N S S TIME SLOTS Buffer Buffer not empty empty
Sensor model • Unknown parameters: • : probability to receive a data unit in a time slot • : probability to transmit a data unit in a time slot • Computes: • Probabilities of phases R,S,N • Average data generation rate • Sensor throughput • Average buffer occupancy
System solution Model decomposition SENSOR MODEL NETWORK MODEL INTERFERENCE MODEL Iterate with a Fixed Point Solution
Network model: assumptions • Each node A maintains up to M routes (according to some routing protocol) • Each route is associated to a different next-hop (a neighbor of A within radio range) • To forward a data unit, node A selects the best next-hop currently available to receive …zzz… Example: M = 3 1 A 2 3
Locally generated traffic (computed by the Sensor Model) Total traffic forwarded by the sensors Routing matrix Network model • The sensor network can be modelled as an open queuing network • The routing matrix is computed according to routing policy of each sensor, and the sleep/active dynamics of its neighbors
System solution Model decomposition SENSOR MODEL NETWORK MODEL INTERFERENCE MODEL Iterate with a Fixed Point Solution
Wireless channel : assumptions • Common maximum radio range r • Ideal CSMA/CA protocol with handshaking (RTS-CTS) • No collisions • No wasted slots • Error-free channel • At each time slot, all feasible transmissions occur successfully • The only constraint is interference (channel contention)
Probability that A can transmit a data unit in a time slot (parameter of the sensor model) Interference model Total interferer Partial interferer G F B C A E I H D
Analysis of data transfer delay • A separate markov chain is build to compute the transfer delay distribution for each sensor node • The state represents the location of a data unit while moving towards the gateway
0.24 mJ/slot 0.24 mJ/slot 0.057 mJ/slot Numerical results N = 400 sensors Radio range r = 0.25 Number of routes M = 6 • Energy consumption: • active mode : 0.24 mJ/slot • sleep mode : 300 nJ/slot • sleep active transition : 0.48 mJ • transmission/reception of data units:
Average transfer delayfor 40 different sensors (p = q = 0.1) 18 sim - load = 0.9 16 mod - load = 0.9 14 12 10 data delivery delay (slots) 8 6 4 sim - load = 0.4 2 mod - load = 0.4 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 distance from sink
Transfer delay distribution for the farthest sensor (p = q = 0.1) sim - load = 0.4 mod - load = 0.4 sim - load = 0.9 0.1 mod - load = 0.9 pdf 0.01 0.001 0 10 20 30 40 50 60 data delivery delay (slots)
sim mod delay [slot] Energy / delay trade-off (1) 0.3 (load = 0.4) sim 100 mod 0.25 energy cons. [mJ] 0.2 0.15 10 0.1 SLEEP 0.05 1 0 p q 0.1 1 10 q/p ACTIVE
sim mod delay [slot] Energy / delay trade-off (2) 30 0.24 (load = 0.9) sim mod 0.22 energy cons. [mJ] 25 0.2 0.18 20 0.16 0.14 15 0.12 10 0.1 0.025 0.05 0.1 0.2 0.4 p ( = q )
Conclusions and future work • We have developed an analytical model of a wireless sensor network, capable of predicting the fundamental performance metric and trade-offs • Many possible extensions: • Introduction of hierarchy (clusters) • Finite buffers and channel errors • Congestion control mechanisms • More details at the MAC level • Impact of node failures network lifetime
References • Carla Fabiana Chiasserini, Michele Garetto, “Modeling the Performance of Wireless Sensor Networks”, IEEE INFOCOM, Hong Kong, March 7-11, 2004
Sensors unfairness 0.011 0.3 energy consumption 0.01 0.25 0.009 0.008 0.2 0.007 0.006 0.15 energy consumption generation rate 0.005 0.004 0.1 0.003 0.002 0.05 generation rate 0.001 0 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 distance from sink
Sensor model validation Average Generation Rate Sensor Throughput 0.003 y = x y = x 0.05 0.0025 0.04 0.03 0.002 mod mod 0.02 0.0015 0.01 0.001 0 sim sim Average Buffer Occupancy Probability of Phase N 1.4 y = x y = x 0.4 1.2 1 0.3 0.8 mod mod 0.2 0.6 0.4 0.1 0.2 0 0 sim sim
Network model validation 1 y = x sensor throughput 0.1 mod 0.01 0.001 0.001 0.01 0.1 1 sim
Interference model validation 1 0.9 0.8 0.7 0.6 β 0.5 0.4 sim 0.3 mod 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 distance from sink
Assumptions - CSMA/CA (RTS/CTS) D …zzz… E F …zzz… B CTS C A …zzz… G RTS
Modern Sensor Nodes UC Berkeley: COTS Dust UC Berkeley: Smart Dust UC Berkeley: COTS Dust Rockwell: WINS UCLA: WINS JPL: Sensor Webs
Interference model validation 1 0.9 0.8 0.7 0.6 β 0.5 0.4 sim 0.3 mod 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 distance from sink
Interference model validation 0.08 sim 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 distance from sink