1 energy-efficiency in manets task 4: energy-efficient networks katia obraczka university of...

50
1 Energy-Efficiency in MANETs Task 4: Energy-Efficient Networks Katia Obraczka University of California, Santa Cruz [email protected] http://inrg.cse.ucsc.edu/

Post on 20-Dec-2015

215 views

Category:

Documents


0 download

TRANSCRIPT

1

Energy-Efficiency in MANETs

Task 4: Energy-Efficient Networks

Katia Obraczka

University of California, Santa Cruz

[email protected]://inrg.cse.ucsc.edu/

2

Summary of Activities

3

Efficient Protocols for Power-Constrained Heterogeneous MANETs

Novel energy-efficient data collection algorithms using spatial and temporal data locality.

Novel flexible interconnection protocol to accommodate device heterogeneity and application requirements.

I. Solis, Efficient Protocols for Power-Constrained Heterogeneous Wireless Ad-Hoc Networks, PhD Dissertation, UCSC, 2005.

I. Solis and K. Obraczka, In-Network Aggregation Trade-offs for Data Collection in Wireless Sensor Networks, International Journal on Sensor Networks (IJSNet), Vol 1, No 2, 2006.

4

Robust Routing for Network Fault-Tolerance and Security (Task 3)

Novel game-theoretic stochastic routing framework as proactive alternative to today's reactive approaches to route repair.

Collaboration with Prof. J. Hespanha, affiliated with UCSB’s ICB (Army UARC) and Prof. S. Bohacek at UDel, funded by the Communications&Networking Army CTA.

Integrated and flexible approach to secure routing in MANETs.

C. Lim, Scalable Multi-path Routing for Robust Communication, PhD Dissertation, USC, 2006.

G. Huang, Robust and Secure Routing in MANETs, MSc Theis, UCSC, 2006. C. Lim, S. Bohacek, J. Hespanha and K. Obraczka, Hierarchical Max-Flow

Routing, IEEE Globecom 2005. S. Bohacek, J. Hespanha, J. Lee, C. Lim and K. Obraczka, A New TCP for

Persistent Packet Reordering, IEEE/ACM Transactions on Networking, Vol. 14, No.2, April 2006.

R. Guru, G. Huang and K. Obraczka, An Integrated and Flexible Approach to Robust and Secure Routing in MANETs, IEEE IC3N, August 2005.

5

Modeling Data Networks with Hybrid Systems (Task 5)

New approach to modeling, analyzing, and simulating networks using hybrid systems which combine both continuous-time dynamics as well as discrete-time logic.

Collaboration with Prof. J. Hespanha, affiliated with UCSB’s ICB (Army UARC) and Prof. S. Bohacek at UDel, funded by the Communications&Networking Army CTA.

J. Lee, A Hybrid Systems Modeling Framework for Transport Protocols, PhD Dissertation, USC, 2005.

S. Bohacek, J. Lee, J. Hespanha and K. Obraczka, Modeling Data Communication Networks Using Hybrid Systems, IEEE/ACM Transaction on Networks, 2006, to appear.

6

Mobility Models for Wireless Networks New approach to modeling mobility in wireless

networks using statistical equivalent models (SEMs).

Collaboration with Profs. B. Sanso and A. Kottas, UCSC Applied Math.

K. Viswanath and K. Obraczka, Modeling the Performance of Flooding in MANETs (Extended Version), Computer Communications Journal (CCJ) 2005.

K. Viswanath, K. Obraczka, A. Kottas, B. Sanso, A Statistical Equivalent Model for Random Waypoint Mobility: A Case Study, IEEE SMC SPECTS 2006.

7

Energy Consumption Modeling, Characterization, and Prediction

Models for energy consumption and network lifetime prediction.

Collaboration with Prof. R. Manduchi, UCSC CE.

C. Margi, Energy Consumption Trade-Offs in Power-Constrained Networks, PhD Dissertation, UCSC, 2006.

C. Margi, K. Obraczka, R. Manduchi, Characterizing System Level Energy Consumption in Mobile Computing Platforms, IEEE WirelessCom 2005, June 13-16, 2005

C. Margi, V. Petkov, K. Obraczka, R. Manduchi Characterizing Energy Consumption in a Visual Sensor Network Testbed, IEEE/Create-Net TridentCom 2006, March 1-3, 2006.

Energy Consumption Trade-offs in Visual Sensor Networks”. C. B. Margi, R. Manduchi , K. Obraczka. SBRC 2006, May 29 - June 02, 2006.

8

Energy-Efficient Medium Access Novel efficient and flexible medium access

framework form MANETs. Collaboration with Prof. J.J. Garcia-Luna.

V. Rajendran, Medium Access Control Protocols for MANETs, PhD Dissertation, UCSC, 2006.

V. Rajendran, K. Obraczka and J.J. Garcia-Luna, Application-Aware Medium Access for Sensor Networks, 2nd IEEE International Conference on Mobile Ad-hoc and Sensor Systems (MASS), November 2005.

9

Energy Consumption Modeling, Characterization, and Prediction

In collaboration with Roberto Manduchi

10

Motivation

Understand energy trade-offs between computation and communication. Why? Make application-level decisions, e.g.,

data processing in the node vs. transmission of all data.

Make resource management decisions, e.g., wake up more often.

Main assumption for sensor nets: communication dominates energy consumption. True?

Heterogeneity in MANETs: Platforms. Sensors. Application requirements.

Mote

Stargate

11

Contributions

Duty cycle energy consumption prediction based on elementary task composition.

Simple lifetime prediction model based on elementary task composition considering different duty cycles.

Case study: visual sensor network.

12

Duty cycle energy consumption prediction

Duty cycle as the node’s “execution unit”. E.g., image acquisition duty cycle.

Duty cycle composed of “elementary tasks”. E.g., capture image, transmit image.

13

Approach: task composition Compose elementary task

consumption to obtain duty cycle consumption. Compose elementary tasks differently for

different duty cycles. Average current does not provide

enough information by itself. Need better granularity:

Charge & duration of a task.

14

Ti: task i. q(Ti): average charge for task i. d(Ti): average duration for task i. Qdc−j: average charge of duty cycle j. Ddc−j: average duration of duty cycle j. n: number of tasks in duty cycle j.

n

1iij-dc )q(T Q

n

1iij-dc )d(T D

Hypothesis

15

First step: task characterization

Thorough energy consumption characterization. Steady state AND transitory behavior.

Case study: Visual sensor network.

16

Visual sensor network node

17

Sensing/Processing: Acquire image; Acquire/save (raw) image; Acquire/compress/save image; Acquire/process image:

No object; Object: then must compress & save sub-image.

Communication: Transmit image:

Raw image (200KB); Full compressed image (48KB); Compressed sub-image (3 different sizes).

Elementary tasks

18

Transitions: Tasks:

Activate/deactivate webcam; Activate/deactivate wireless card; Sleep/wakeup.

Elementary tasks

19

Elementary tasks: duration

0.0000

0.2000

0.4000

0.6000

0.8000

1.0000

1.2000

1.4000

1.6000

Du

rati

on

(s)

Results are the average of 20 different measurements.

20

0.0000

0.0200

0.0400

0.0600

0.0800

0.1000

0.1200

0.1400

Incr

emen

tal

Ch

arg

e (C

)

Elementary tasks: charge

21

Duty cycles 6 different duty cycles. 2 types:

Deterministic: image acquisition/compression.

Conditional: event detection. If no event is detected, the system is put in

sleep or idle mode for T1 = 5 seconds. Otherwise, the system remains idle for T2 =

3 seconds.

22

Duty Cycle (b)Duty Cycle (c)Duty Cycle (a)

Image

Tx

Wait

Image

Tx

Wait

Deterministic duty cyclesIm

age

Tx

Wait

23

Conditional duty cycles

Duty Cycle (d) Duty Cycle (e) Duty Cycle (f)

Imag

e

Tx

Wait

Imag

e

Tx

Wait

Tx

Wait

Imag

e

Yes Yes Yes

24

Duty cycles: duration & charge

0

2

4

6

8

10

12

14

16

(a) (b) (c) (d) - noevent

(d) -event

(e) - noevent

(f) - noevent

(f) -event

tim

e (s

)

Predicted

Measured

0

0.5

1

1.5

2

2.5

3

3.5

4

(a) (b) (c) (d) - noevent

(d) -event

(e) - noevent

(f) - noevent

(f) -event

Ch

arg

e (C

)

Predicted

Measured

Relative error:Ed: -6.6% to 3.4%.

Eq: -9.4% to 9.1%.

Results are the average of 20 measurements.

predicted

measuredd D

DE 1

25

Errors Webcam & wireless card activation:

Must add 1 second delay after in the duty cycle scripts after issuing commands!

But it is not 1 second in idle! Hardware still working!!

Linux as the OS: Does not allow complete control.

Wireless interference: TX task has variable duration; Transmission errors.

26

Lifetime prediction

27

Simple lifetime prediction model Follows same hypothesis. Lifetime:

Lx: node lifetime. Qb: charge available from the battery. Qd: average charge for each duty cycle. Dd: average duration of each duty cycle.

dd

bx D *

Q

Q L

28

Simple lifetime prediction model

Expand formula to include duty cycle prediction:

For conditional duty cycles:

n

1iin

1ii

bx )d(T *

)q(T

Q L

)P - (1 * Q P * Q )E(Q ev-dno-dev-dev-dd

29

Experiments Same duty cycles presented. Duty cycle runs continuously, until

1000 mAh is used. No control on event generation (node

was placed in our lab).

30

Simple lifetime prediction model

Prediction vs. Experiments

Relative error for prediction based on duty cycle model is less than 13%.

Results are the average of 20 measurements.

0

5000

10000

15000

20000

25000

(a) (b) (c) (d) (e) (f)

Lif

etim

e (s

)

Based on DC measurements

Based on DC Prediction

Measured Lifetime (s)

Duty cycle duration:

(a): 6.4 s

(b): 11.6 s

(c): 13.9 s

(d) - no object: 11.0 s

(e) - no object: 8.3 s

(d) & (e) – object: 9.5 s

(f) - no object: 6.2 s

(f) - object: 4.4 s

31

Lifetime prediction model:summary

Simple. Relative error < 13%. Allows lifetime estimation for new

deployments. Allows duty cycle trade-off analysis.

32

Future Directions (1)

Formalize lifetime prediction model to include non-deterministic sequence of tasks: Set of known tasks. Sequence of tasks not known a priori. Inputs: info from neighboors, battery, etc. Could we use it to implement a “resource

manager” on the node?

33

Future Directions (2)

Validate this approach using different hardware platforms and sensor network applications. Would a simpler hardware platform allow

better accuracy?

34

Energy-Efficient Channel Access

In collaboration with J.J. Garcia-Luna

35

Contributions

First traffic-adaptive MAC protocol. TRAMA.

Application-aware MAC. FLAMA (Motes testbed). MFLAMA.

Framework for energy-efficient, application-aware, multi-channel medium access. DYNAMMA (UWB radio testbed).

36

MAC state-of-the-art

Far from addressing challenges posed by: Taking advantage of higher PHY data

rates. Accommodating different applications. And still achieve energy efficiency.

37

Approach: DYNAMMA (Dynamic Medium Access Framework) Flexible, energy-efficient, application-aware

framework. Flexible and efficient traffic announcement

mechanisms. Slot structure with reduced idle duration and

inter-frame spacing. Scheduled access (including signalling).

Avoids collisions. Multi-channel.

Spatial re-use for improved channel utilization.

38

Time slot organization

Superframe NSuperframe N+1

Signaling Slots

Burst Data Slots

Base Data Slots

Collision-free signalingGather neighbor Information

Collision-free data exchangeBurst data frame exchange

Collision-free data exchangeSingle frame exchange

39

DYNAMMA: components

Collision-free signalling.

Traffic characterization. Different class of flows

based on flow arrival / service rate.

Each flow class contends for a “subset” of the channel access slots – prevents idle slot allocation.

Multi-channel, collision-free scheduling. One transmitter per

channel in the 2-hop. Channel selection based

on flow priorities. Ensures that a node

does not transmit to a node sleeping or listening on other channels.

40

DYNAMMA performance Performance analysis by extensive

simulations using QualNet and testbed experiments using UWB radio/MAC platform.

Different application scenarios Synthetic – random exponential traffic (worst

case scenario for DYNAMMA, large number of flows).

Data gathering application.

41

Synthetic Traffic

42

Simulation Setup

16 nodes, square grid topology (18m). 4000 bytes packet. DYNAMMA Parameters:

SignalingSlot = 16. BaseSlots = 16. BurstSlots = 240, framesPerSlot = 2.

43

Delivery Ratio

Packet losses forDYNAMMA and TRAMAdue to queue drops.

Packet losses for 802.11due to collisions.

Multiple channels helpreducing the queuingdelay.

44

Queuing Delay

45

Energy savings

DYNAMMA implementsidle receive timeouts -i.e. shut off receiver if thetransmission does not startwithin a timeout.

Idle receive timeoutsimproves energy savings.

46

UWB Testbed

Instruction / Data BRAM64 KB

PLB Controller

PowerPC 405

DCR Interface

PLB Controller

Processor Local B

us

DC

R B

us

Master PLB

Master DCR

PLB to OPB Bridge

OP

B B

us

RS232 Interface

FPGA Setup

clk

rst

Lower MAC

MPI

RS232TX

RS232RX

Radio Mode Control Schedulers

Timers

Transmit / receive DMA

47

Testbed experiments

Basic slot duration: 644 us Signal slot duration: 161

us Superframe: 16 signal

slots, 256 base slots = 167.440 ms

SIFS = 10 us, MIFS = 1.875 us

Time base using a 66.66 Mhz crystal with < 10ppm drift. Total drift per superframe 1 us.

Three nodes – saturated throughput analysis

Two data rates – 53.3 Mbps, & 200 Mbps 4000 bytes payload for

53.3 Mbps 3400 bytes payload and

“4” burst per slot for 200 Mbps

49

Experimental results

53.3 2000

5

10

15

20

25

30

35

40

Per-flow Throughput

A-BA-CB-AB-CC-AC-B

Data rate (Mbps)

Thro

ughput

(Mbps)

A

C

B

53.3 2000

10

20

30

40

50

60

70

80

90

100

Sleep / Utilization

Sleep

Utilization

Data rate (Mbps)

Perc

enta

ge

50

Conclusion

Flexible framework for energy-efficient, application-aware medium access.

Significant improvements in delay and reliability.

Significant improvements in channel utilization by the use of multiple channels.

51

Future Work

Improve traffic characterization using prediction.

Scheduling optimizations to increase the number of schedules. Trade-off correctness? Deal with inconsistencies.

Alternate scheduling approaches to provide guarantied QoS.