Transcript ppt

An Adaptive EnergyEfficient MAC Protocol
for Wireless Sensor
Networks
Tijs van Dam
Koen Langendoen
Presenter: Michael Curcio
ACM SenSys 2003
Sensor Networks
•
•
•
•
•
Low message rate
Insensitive to latency
Low processing power and memory capacity
Lots of redundancy
Often battery operated!
Goals
•
•
•
Focus has moved away from maximizing
throughput and fairness; minimizing latency
Power consumption kept to a minimum
Memory/Network processing kept low
Energy Sinks
•
•
•
•
•
Processor
Radio
•
•
Receiving/Transmitting
Idle Listening
Collisions
Protocol Overhead
Overhearing
What else addresses this?
•
•
•
•
•
•
TDMA
802.11 (CSMA)
Extra wake-up radio
TinyOS
S-MAC
Radio-triggered wake-up hardware
How do they do that?
•
•
TDMA
•
•
Built-in duty cycle; eliminates collisions
But-- Hard to do for ad-hoc network
•
•
Scheduling, slot allocation, coordination
Clock drift (especially for small slots)
Extra radio
•
•
Solve problems with more hardware
Add a radio on a different frequency that can
wake up other nodes
How do they do that?
•
802.11
•
•
•
Has power-saving features, but not good
enough for sensor networks
Was designed for nodes all existing in one
cell (no multi-hop)
TinyOS
•
•
Instead of listening really long for a short
transmission, listen realy short for a long
transmission
Makes the transmitter, not the receiver pay
the energy bill
How do they do that?
•
S-MAC
•
•
•
•
Fixed duty cycle
Compress spread out transmits and receives
into a shorter amount of time so we can
sleep the rest
Event, issue, request-based transfer of
information hop-to-hop
Radio-triggered wake-up
•
•
Stay awake for Soji’s presentation
Teaser: On-demand node wake-up
T-MAC Approach
•
•
“Timeout”-MAC
Adaptive duty cycle
•
•
•
Period of radio activity can be ended
dynamically
Reduce idle listening to a minimum
Better handle variable network load
Sensor Network
Communication Patterns
•
•
Local uni-/broadcast
•
Event processed in network among nodes
Nodes to sink(s) reporting
•
•
Messages move through the network in a
generally unified direction (to the sink(s))
May or may not be aggregated/processed
en-route
Sensor Network
Communication Patterns
•
•
•
Traditional, multi-hop routing not used
•
Might there be a case where it would be
useful?
Time dependence
•
Nothing to do if no events occur
Location dependence
•
Network in proximity to sink nodes
experiences heavier traffic than at remote
edge of network
EYES Nodes
•
•
•
•
•
•
16-bit, 5MHz,
variable clock rate
processor
2KB RAM
60KB FLASH
2MB EEPROM
JTAG, RS232, 2
LEDs, 16 GPIO (8
ADC) pins
Runs on 2 AA
batteries
Photo Courtesy: Eyes - Energy Efficient Sensor
Networks,
http://www.eyes.eu.org/
T-MAC Duty Cycle
•
•
•
•
Variable length duty
cycle
Transmit in bursts
Maintain optimal
active time under
variable load
Sleep after time of
hearing nothing
S-MAC Duty Cycle
•
•
•
Fixed duty cycle
Frame time - limited
by latency
requirements and
buffer space
Active time configured to be
long enough to
handle highest
expected load
T-MAC Protocol Basics
•
•
•
•
Burst communication schedule
Messages are queued while node is asleep
•
Buffer capacity determines Frame Time
RTS/CTS/<Data>/ACK
•
•
Collision avoidance
Reliability
Active Period (Active Time)
T-MAC Active Period
•
•
Starts at scheduled intervals
Ends when no Activation Event is heard for a
time = TA
•
•
•
•
•
firing of a periodic frame timer
reception of any data
sensing of any communication activity
end-of-transmission of own data or
acknowledgment
overhearding end of neighbor’s data
exchange
T-MAC Considerations
•
•
•
•
Clustering and Synchronization
RTS Operation and Choosing TA
Overhearing Avoidance
Asymmetric Communication
Clustering and
Synchronization
•
•
From S-MAC protocol
Virtual Clustering
•
•
Frame schedules and SYNC packets define
a node’s active time
Shared with neighbors to ensure
transmissions go to nodes that are awake
RTS and TA
•
•
•
Fixed contention
window at
beginning of active
time for RTS
signaling
RTS retry after loss
(max of two times)
TA > C + R + T
Overhearing Avoidance
•
•
Results in energy savings, but decreases max
throughput (do not use if speed is required)
Experiments show going to sleep to avoid
overhearing makes nodes miss other
RTS/CTS transmissions. When they wake-up,
they cause interference collisions.
Asymmetric Communication
•
•
Most
communication is
unidirectional
(node-to-sink)
Early Sleeping
Problem
•
•
Future-Requestto-Send (FTRS)
Full-buffer Priority
•
•
•
•
Future RTS
Nodes that lose
RTS/CTS
contention have
opportunity to send
FRTS
Collides with empty
DS packet of
contention winner
Requires increase
in TA (increased
energy usage)
75% throughput
gain
Full-buffer Priority
•
•
When a node’s
sender buffer is or
is almost full, it can
decide to ignore an
incoming RTS, i.e.,
refuse to send a
CTS reply, and
sends its own RTS
to a different node
But... heavy load
increases collisions
rapidly
Experimental Setup
•
•
•
•
•
OMNeT++ discrete event simulation package
EYES nodes modeling
100 nodes on 10 x 10 grid; non-edge nodes
have 8 neighbors each
Local Unicasts
Nodes-to-sink communication; shortest-path
routing
Results
•
•
Simulations
•
•
•
•
Homogenous local unicast
Nodes-to-sink communication
The effects of early sleep
Event-based local unicast and node-to-sink
reporting
Real Implementation
•
Energy use
Homogenous Local Unicast
Nodes-to-sink
Communication
Early Sleeping
Event-based Unicast and
Node-to-sink
Energy Consumption
Future Work
•
•
•
Experimentation with FRTS and full-buffer
priority to solve early sleeping problem
Node mobility
Virtual clustering and multi-hop
synchronization
Conclusions
•
•
•
Power consumption reductions achieved
As much as 96% with low loads as compared
to traditional protocols
Improves upon S-MAC performance in volatile
environments where message rates change
with both time and place