Medium Access Control With Coordinated Adaptive Sleeping for Wireless Sensor Networks Debate 1 - Defense Joseph Camp Anastasios Giannoulis.

Medium Access Control With Coordinated Adaptive Sleeping

for Wireless Sensor Networks

Debate 1 - Defense

Joseph Camp

Anastasios Giannoulis

Problem Statement

Wireless sensor nodes have vastly different energy requirements than 802.11-type nodes Require long periods of being idle Less stringent demands on per-node fairness,

latency, and throughput Network has global objective in contrast to selfish

objectives Need for scalability (unlike TDMA)

Energy Waste

802.11 and TDMA MACs Sources of Energy Waste

Idle listening (50%-100% of energy consumed for receiving)

Collision/Retransmission Overhearing Control Packet Overhead

Need for redesign of MAC for sensor networks -- Sensor-MAC (S-MAC)

Existing MACs

Contention-based 802.11 high energy consumption [SK97] PAMAS reduces overhearing, but needs 2 radios

Does not reduce idle listening TDMA-based (i.e. Bluetooth and LEACH)

Not scalable--hard to change frame length and time slot assignment

Low bandwidth utilization (FDMA, CDMA) Adaptive rate control for fairness (not objective here) Piconet had low-duty-cycle operation, but not coordinated Power-Save (PS) 802.11 for single hop networks

Energy Reduction Techniques

Periodic Listen and Sleep Coordinate sleep schedules Maintain synchronization

Collision/Overhearing Avoidance Physical carrier sensing Virtual carrier sensing (RTS/CTS and keep track

of NAV) Overhearing avoidance Message passing--fragment bursts

Technique 1:Periodic Listen and Sleep

Each node has schedule table about neighbors Node first listens for synchronization period

If hears neighbor’s schedule, follows that schedule Else it sets its own schedule; If it then hears a schedule

One schedule -> follows new schedule Multiple schedules -> adopts all schedules

Technique 1:Periodic Listen and Sleep

Maintaining synchronization Counter clock drift by…

Relative time stamps instead of absolute Receiver subtracts transmission time Listening period >> clock drift

Adaptive listening Wake at end of neighbor’s tx to forward data Known from control packets

Technique 1:Periodic Listen and Sleep

Latency Analysis Common delays to both 802.11 and S-MAC

Propagation and processing (ignored) Backoff and Queueing (none for light traffic) Carrier Sense and Transmission

SMAC-specific delay -- sleep delay Average Latency

Without sleeping With sleeping, without adaptive listening With sleeping and adaptive listening

Technique 1:Periodic Listen and Sleep

Without sleeping (tcs + ttx)

Sleeping, no adaptive listening (Tf)

Sleeping and adaptive listening (Tf / 2)

* Tf >> (tcs + ttx)

Technique 2: Overhearing Avoidance

Overhearing Avoidance After RTS directed for another, go to sleep Neighbors of both sender and receiver

Message Passing Technique to transmit long message Transmitting entire message high probability of

corruption/wasted air time -> wasted energy Burst fragments of packet after one RTS/CTS

(less control overhead)

Protocol Implementation Initial Implementation on

Rene Motes 802.11-like MAC S-MAC without sleep S-MAC with sleep

Current Implementation on Mica Motes Duty-cycle selection Fully active mode Disable adaptive listen Modes:

Rx (14.4 mW), Tx (36 mW) and Sleep (15uW)

Results: Measurement of Energy Consumption

Two-Hop Network 802.11 (no sleep) S-MAC (no sleep)

Overhearing avoidance and message passing

S-MAC (with sleep - 50% duty cycle)

802.11 uses > twice the energy as S-MAC

Periodic sleep gains at greater than 4 s

Rene Motes

Results: Measurement of Energy Consumption

No sleep vs. sleep similar result for multihop

Mica Motes Adaptive listen not as

beneficial for energy savings as latency…

Results: Adaptive Listen and End-to-End Latency

Save equal energy with 10% duty cycle (previous slide) yet achieve performance close to no sleep

1/5 latency with adaptive listening Without adaptive

listening message has to wait one sleep cycle each hop

Results: More on Latency

Similar results for average packet latency for highest traffic load

Highlights variance of latency: Adaptive Listen reduces variance of latency

Results: E2E Throughput

Less throughput than no sleep 10% duty w/ adaptive

listen - 1/2 10% duty w/out adaptive

listen - 1/8

Results inversely hold for throughput At 10 s inter-arrival

period, all schemes have enough throughput

1/21/8

Energy Time Cost Per Byte

All factors incorporated with Energy-time cost per byte

10% duty cycle with adaptive listen performs best for all traffic loads

Heavy Load (< 4 s) Work with duty cycles

Unique Contributions

Implement low-duty-cycle scheme for multihop sensor network

Employ adaptive listening to reduce latency In-channel signaling to avoid energy waste Message passing to reduce control overhead Measurement and evaluation of S-MAC

(tradeoff of energy, latency, and throughput)

Summary

Well-written, well-motivated paper Identify objective, origins of problem, relate to

existing work, and contribute solutions to the identified problems

Combine theory and practice to validate solution Clearly define where the solution would have the

most benefit Neutralize tradeoff between energy, latency, and

throughput

High Impact

Motivated work in the future 107 citations in three years!

First released as a technical report in September, 2002 and submitted to journal in January, 2003

Additional work in MAC for sensor networks (pioneering paper for sensor networks) Ten of the papers that cited this work have 9 or more

citations, one of which has 71 citations INFOCOM’02 Paper (almost same text) has 638 citations

30 papers that cited this work w/ 50 citations or more!