Z-MAC: a Hybrid MAC for Wireless Sensor Networks

Topology Control
Presenter: Ajit Warrier
With
Dr. Sangjoon Park (ETRI, South Korea),
Jeongki Min and
Dr. Injong Rhee (advisor)
North Carolina State University Networking Lab
http://netsrv.csc.ncsu.edu
Introduction: Topology Control
Topology Control/Clustering
■ Reduce structural complexity in a network.
■ Delegate complex/energy consuming
activities to a subset of nodes in the network.
Topology Control Approaches
Power Control
• Most often used in wireless ad-hoc networks.
• Reduce routing complexity.
• Reduce wireless interference.
• Preserve network capacity ? Connectivity ?
Topology Control Approaches
Connected Backbone
B
A
• Most often used in wireless ad-hoc networks.
• Reduce routing complexity.
• Reduce wireless interference.
• Preserve network capacity ?
Topology Control Approaches
Clustering/Hierarchy
• Most often used in wireless sensor networks.
• Reducing complexity not the issue, radio power
consumption is !
• Reduce radio transmissions/energy consumption.
• Do not care (as much) about capacity.
Topology Control – Pros/Cons
Pros
■ Energy Efficient – Radio draws order of magnitude
more energy than the sensing board.
■ Less radio interference.
■ Less routing complexity.
Cons
■ Loss of routing selectivity.
■ Topology maintenance overhead.
Motivation
Lots of theory/simulation – very few experimental
results.
■ Complicated algorithms.
■ Assumptions in the algorithm difficult to realize
in practice:
■
Wireless links usually vary in quality over time.
■
Wireless links not binary in nature.
■
Wireless links may be asymmetric.
■
Sensor nodes have low speed CPUs, may not be
possible to run complex algorithms.
HEED experimental testbed
G2
Mica2 nodes
G1
barrier
G3
Mica2Dot nodes
observer
FLOC experimental testbed
Algorithm and Analysis
Our Topology Control Algorithm - Overview
■ Divide the sensor network into approximately equal regions
called clusters.
■ Cluster Members

Every node belongs to one cluster.

Perform sensing, if an event occurs, transmit event to cluster
head.
■ Cluster Head

Within radio range of all nodes of a cluster.

Responsible for two activities:
 Collect sensing reports from members.
 Route/forward sensing reports toward the sink.
■ Gateways

Member nodes acting as connecting link between two
clusters.
Algorithm - Overview
Cluster Head Election Algorithm
Time-line of a node, in rounds
Cluster Head Election Algorithm
Flip coin with
probability p0
Time-line of a node, in rounds
Cluster Head Election Algorithm
Lose
Flip coin with
probability p0
Time-line of a node, in rounds
Cluster Head Election Algorithm
Lose
Flip coin with
probability p0
Flip coin with
probability kp0
Time-line of a node, in rounds
Cluster Head Election Algorithm
Lose
Flip coin with
probability p0
Lose
Flip coin with
probability kp0
Time-line of a node, in rounds
Cluster Head Election Algorithm
Lose
Flip coin with
probability p0
Lose
Flip coin with
probability kp0
Flip coin with
probability
k2p0
Time-line of a node, in rounds
Cluster Head Election Algorithm
Transmit Cluster Head Announcement (CHA)
Lose
Flip coin with
probability p0
Lose
Flip coin with
probability kp0
Win – Become Cluster Head
Flip coin with
probability
k2p0
Time-line of a node, in rounds
Cluster Head Election Algorithm
Lose
Receive CHA – Become Member Node
Flip coin with
probability p0
Time-line of a node, in rounds
Cluster Head Selection
Gateway Selection
Routing Phase
Data Transmission – Differential
Duty Cycling
• Cluster heads, gateways responsible for
routing/data forwarding => set radio to
high duty cycle.
• Member nodes only responsible for
sensing => set radio to low duty cycle
(ideally to 0%).
• Ratio of duty cycle of member nodes to
that of cluster heads/gateway nodes
decides energy efficiency of network.
Analysis Result – Energy Saving
Ratio d = 0
Ratio d = 0.2
Ratio d = 0.4
Ratio d = 0.6
Ratio d = 1
Topology Control Operations
Experimental Results
Experimental Platform
The algorithm has been implemented on Mica2 sensor
nodes running the TinyOS event-driven operating system.
 Platform:
• Motes (UC Berkeley)
• 8-bit CPU at 4MHz
• 128KB flash, 4KB RAM
• 916MHz radio
• TinyOS event-driven
Experimental Testbed
■ 42 Mica2 sensor motes in Withers Lab.
■ Wall-powered and connected to the Internet via Ethernet ports.
■ Programs uploaded via the Internet, all mote interaction via
wireless.
■ Links vary in quality, some have loss rates up to 30-40%.
■ Asymmetric links also present.
Experimental Testbed – Connectivity
Experimental Testbed – Snapshot
Implementation Details
■ MAC Layer – B-MAC


CSMA-based.
Duty Cycled.
■ Routing Layer – Mint


DSDV-like table driven, proactive
Uses link level measurements to select routing parents.
■ Member nodes switch off their radio. (δ = 0)
■ Cluster heads tested with varying duty cycles (X = 2% - 45%)
■ Radio is 19.2 Kbps, packet payload of 36 bytes.
Experimental Method
■ Every node transmits packets with probability α% per second.
■ α varied for two types of scenarios
 Low Data Rate Experiment
 Nodes idle most of the time, brief periods of activity, e.g.
Earthquake detection.
 α = 0.1 – 1
 High Data Rate Experiment
 Application scenarios with more periodicity, e.g.
Temperature monitoring.
 α = 10 – 100
Algorithm Overhead
■ Total energy of 5 J is 0.03% of the total battery capacity.
■ Half the time overhead is because of routing.
■ Given time synch period of 10s, it is feasible to use a
reclustering period of 17 hours.
Energy Efficiency – Low Data Rate
Topology Control
2% Duty Cycle
5% Duty Cycle
B-MAC
10% Duty Cycle
Energy Efficiency – High Data Rate
Topology Control
2% Duty Cycle
5% Duty Cycle
B-MAC
10% Duty Cycle
Throughput
B-MAC
B-MAC
Topology
Control
Topology
Control
Conclusion and Future Work
■ As a thumb rule, topology control can extend network lifetime by
the network density divided by 4-8.
■ Topology control is not necessarily capacity conserving, may
result in up to 50% loss in throughput. This is due to reduced
routing selectivity.
■ Given the mathematical analysis, one may attempt to optimize
the algorithm for some system performance metric, for instance
throughput.
■ Need to develop robust algorithms for node failure resolution.