Johns Hopkins University Department of Computer Science Graduate Board Oral Presentation May 23 2012 Advisor Dr Andreas Terzis Chair Dr Alex Szalay Doug Carlson Outline Background Problem statement ID: 789798
Download The PPT/PDF document "Hardware, Software, and Protocols for th..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Hardware, Software, and Protocols for the next generation of Environmental Monitoring Networks
Johns Hopkins University Department of Computer ScienceGraduate Board Oral PresentationMay 23, 2012Advisor: Dr. Andreas TerzisChair: Dr. Alex Szalay
Doug Carlson
Slide2Outline
BackgroundProblem statementThe Breakfast hardware suiteCX: leveraging concurrent transmissions for multipath data collection
2
Outline
Slide3Background
Background. Problem Statement. Hardware. CXBackground3
Slide4WSNs for
Environmental MonitoringSample periodically (~O(minutes))Collect data wirelesslyLoose latency
requirements (hours or days)
High yield
requirements (>99%)
Survive on batteries for months or years
4
150 m
Background
Slide5Current Approach
HardwareTelosB “Mote”Microcontroller, radio, storageUp to 4 analog inputsSoftwareKoala [Musaloiu et al, 2008]Radios off except for batch downloads
Centralized source routing
Offline timestamp reconstruction [EWSN 2010]
Delta compression, storage stack [REALWSN 2011]
Background
5
Slide6Problem statement
Background. Problem Statement. Hardware. CXProblem Statement6
Slide7Larger Networks: More Data Forwarding
Problem Statement7
Slide8Standard Protocols Don’t Scale Well
Problem Statement8Source: Does Wireless Sensor Network Scale? A Measurement Study on
GreenOrbs
. Liu et al, INFOCOM 2011
Slide9Route Selection
is Hard9Problem StatementHard-to-capture effects present
Link quality freshness matters
Figure 1
Slide10Problem: Technical Approach Doesn’t Match Usage Patterns
10High deployment and maintenance effortEnergy usage and routing failures grow with network size
New hardware: efficient allocation, automate tasks.
Hierarchical networks, robust multi-path data collection
Problem Statement
Problem
Solution
Slide11Hardware
Background. Problem Statement. Hardware. CXHardware11
Slide12Matching Hardware to
Deployment Patterns12HardwareFigure 4
Slide1313
More efficient allocation
Automate tracking tasks
Decoupling
: energy savings, SW design
improvements
Hardware
Figure 4
Slide14Reduce Deployed and Maintained Hardware
USDA deployment20 TelosB nodes Replaceable with 2 Bacon + 10 Toast<50% of HW cost1/10th of the monitored/maintained devices14Hardware
Node
Relay
Uplink
Slide15Eliminate Manual Tracking Tasks
15Hardware
Field
Database
Lab
Data
Calibrate
Toast
Register sensors
Sensor associations
Record sensor associations
Enter sensor associations
Deploy
Deploy
Slide16Hardware Status
Hardware is in productionDrivers written for peripheralsToast auto-discovery and metadata storage written End user analog sensor setup Integration with existing data pipeline16Hardware
Slide17CX
Background. Problem Statement. Hardware. CXCX17
Slide18Routing Revisited
Hardware alleviates bottleneck nodes and reduces network depth.Route selection and maintenance is still hard in low-power setting.Route-free data collection: possible with network floodsNetwork floods can be made efficient with concurrent transmissions [Ferrari et al, 2012]18CX
Bacon hardware lets us schedule concurrent transmissions better than existing hardware does.
This lets us build more efficient techniques than flooding.
Slide19Slot
Cycle
0
1
2
…
100
101
102
…
5000
5001
5002
…
…
Frame
CX Principles
In a given frame, all nodes transmitting a packet transmit the same packet.
Each slot
is “owned” by one node.
In a given frame, all nodes which transmit do so at the start of the frame.
19
Single
TDMA schedule
, disseminated by root
node
CX
Slide20Flood
CX20
Slide21Scoped Flood
One-to-one communication2 ACK frames to every Data frameAcks “catch up” and suppress flood.21
CX
Figure 6
Slide22Scoped Flood
CX
22
Slide23Scoped Flood
Learn distance, establish forwarders
Send first packet via
(pre-routed) flood
Delay based on distance
D
…
A
…
F
…
…
…
F
…
…
…
…
Send next packet
Transport Layer- Unreliable Burst
23
CX
Slide24Initial Results: Link/
Phy Layer24CXFigure 7(a)
Figure 7(b)
Slide25Initial Results:
Network/Transport LayerPower: 7.2mW (16% of always-active)Average goodput: 17.6B/min/nodeCompetitive with Koala when corrected for goodput25
CX
Figure 8
Slide26Improving CX Performance: Delivery
Forward Error CorrectionMeasure/validate burst packet spacingInclude shortest paths +1 in burstIncorporate retransmissions26CX
Figure 7 (b)
50K
100K
125K
250K
FEC Off
96.2%
75.2%
88.8%
11.9%
FEC On
99.5%
95.9%
94.9%
15.8%
Slide27Improving CX Performance: Duty
CycleIncorporate transport layer information27CXNo burst transfer detected
Not on shortest path
One Slot
Sender
On shortest path
TX
RX
Off
Slide28Possible Expansions
Adapting TX power and symbol rate to trade energy for rangePerformance in high-mobility scenariosApplications to relay placement in partially-connected areasCX28
Slide29To Recap
Problem: Mismatches between hardware, networking, and real-world deployment patterns limit scalability.Breakfast hardware suite: reduce cost, ease deployment effort, improve maintainability.Hierarchical network structure: limit impact of deployment size on sensing node lifetime.In-patch data collection: Energy-efficiency and robustness possible with CX.29Conclusion
Slide30Acknowledgements
Andreas Terzis, Kathy Szlavecz, and Alex Szalay for guidance and motivation.Razvan M-E, Jayant Gupchup, Mike Liang for laying the foundations for our work in EM.Marcus Chang for help/guidance on the intricacies of developing for new platforms.Yin Chen and Marcus Chang for nurturing the ideas in CX, running experiments, and generally working really, really, really hard.Scott Pitz, Lijun Xia, Chih-han
Chang, Mike Bernard, and many others that made deployments possible.
The GBO Committee for your time in a very busy day.
30
Conclusion
Slide31Current Approach- Software
KoalaSample/store data periodicallyCentral basestation wakes up networkBasestation uses local network connectivity information to build source-routed paths to each nodeDownload outstanding data from each node, one at a timeReturn network to sleep31Background
Slide32WSNs for Environmental Monitoring
Wirelessly collect sensor data.Sensors at sites of interest.Survive on batteries for months or years.
32
Background
Slide33Table 1: Summary of Deployments
Background33
Slide34Why is Bacon better at this than Telos?
Access to accurate and precise high frequency timer (26 MHz)Flexible symbol rate (< 2 Mchip/S on Telos)Permits delays of many mS between receiving and retransmittingLacks coding redundancy (e.g. Direct Sequence Spread Spectrum)34
CX
Slide35Principle of concurrent transmissions
35CX
Slide36CX Network Stack
36CXFigure 5
Slide37Figure 6: Flood v. Scoped Flood
CX37
Slide38Flood
One-to-many(Ideally) reaches all nodesCompletion time bounded by network diameter38
CX
Figure 6
Slide39Figure 7: Single-hop CX tests
CX39
Slide40Eliminate Manual Tracking Tasks
40
Enter Metadata
Record sensors and associations
Store Calibration Results
Upload Metadata
Record sensors
Hardware
Slide41Decouple sensing, storage, and communication
Reduce powerCentralize energy consumptionImprove SW design41Toast
Sensing, compression, signal processing
Bacon (Leaf)
Storage, forwarding within patch
Bacon (Router)
Storage, forwarding to sink
TelosB
Node
Sensing, compression, signal processing,
storage, forwarding to sink
Energy Consumption
Hardware
Slide42Table 2: Hardware Comparison
Hardware42
Slide43Figure 4: Breakfast Overview
Hardware43
Slide44Coping with Disconnection: Phoenix (EWSN 2010)
44Use set of time references to map mote measurements to global time scaleIncreased yield from < 90% to >99%
Previous Work
Figure 2
Slide45Adapting to Campaign Deployments (REALWSN 2010)
Delta compression: 70-74% savings vs. uncompressed>99% of total data yield timestamped
45
Previous Work
Figure 3
Slide46Flip-MAC: Density-adaptive contention reduction (DCOSS 2011)
Previous Work46
Slide47Flip-MAC results
Previous Work47
Slide48Impact of instability on route success
Problem Statement48
Slide49Intentional well-formed collisions
CX- Expansion49
Sender 1
Sender 2
Receiver
Slide50Relay clustering
CX- Expansion
50
Slide51Robust one-to-many communication
Disseminate: flood data packetRepair:Alternate NACK and DATA framesNodes without packet send NACK during NACK frameNodes with packet send data if they got a NACK in the previous frame. CX- Expansion51
Slide52CX and Mobility
No persistent routing statePath discovery completes in << 1 secondCX- Expansion52
Slide53Adaptive power and data rate
Lower symbol rate: better sensitivityPower consumption constant, duration changesReduce network depthCX- Expansion53
Slide54Related work: ExOR (
Biswa & Morris, 2005)Attempts to opportunistically exploit long, lossy linksBroadcast, forwarder chosen from pool of receiversChallenge is to get pool of receivers to agree on which is the “closest” to the destinationGoal is to improve throughputRelated Work54
Slide55RW: Zigbee
Router nodes may not duty cycleStar network: single hop onlyTree network: static routingMesh network: AODV, source routingRelated Work55
Slide56Related Work- Analaysis of multipath routing Part I: The effect on the packet delivery
ratio (Tsirigos and Haas, 2004)Motivated by link instability in wireless ad hoc networksSplit up/encode data with some redundancySend encoded chunks on different pathsDistributes number of chunks on each path based on that path’s estimated PRR.Related Work56
Slide57RW: Fully Wireless Implementation of Distributed
Beamforming on a Software Defined Radio Platform (Rahman, Baidoo-Williams 2012)Beamforming: attempt to phase-align transmissions to maximize constructive interference at receiverAccomplished using simple software techniques: senders randomly perturb phase of their transmission, get feedback on whether it increased or decreased RSS from receiverNot applicable to flooding: we want to reach multiple physical locations so we can’t use beam-formingWould be nice to have some support for this in future radio HW
Related Work
57
Slide58RW: The Capacity of Wireless Networks (Gupta & Kumar, 2000)
Under optimal conditions, throughput for a node is theta(w/sqrt(n)), w = speed n = nodes in networkAt a high level: throughput diminishes as the nodes in a network increaseTry to keep networks small!Related Work58
Slide59RW: Glossy (Ferrari, Zimmerling, Thiele et al)
Efficient Network Flooding and Time Synch with Glossy (IPSN 2011)The Bus Goes Wireless … (IQ2S 2012)Also advocates a routing-free communication strategyApproaches lower bound on flood latency60 second IPIAchieves > 99% end-to-end PRRDuty cycle < 2% on 85-node 3-hop testbedOur approach has same number of transmissions for flood, fewer for scoped flood. Outperforming this will come down to duty cycle tuning. Related Work
59
Slide60(7,4) code: Rate = 0.5 when including 2-bit error detection
implemented with relatively small lookup tables, higher coding rates require code/decode routinesRelated Work60
RW: Hamming Codes
(Source: Wikipedia)
Slide61RW: Trickle (Levis et al, 2004)
“Polite gossip” for disseminating small units of data throughout a networkNodes adaptively increase the delay between retransmitting a packet when it’s been “in the air” for a whileExemplifies the difficulty of coordinating floodsRelatively long propagation delays (10s of seconds/minutes) Related Work61
Slide62RW: Flash Flooding (Lu, Whitehouse 2009)
Rely on capture effect to survive collisionsAttempts to control number of concurrent transmissions, detect/restart failed floodsEdges of network see poor completion timesNote Naïve Flood completion (X-MAC)Related Work
62