networked embedded systems and the zebranet project: … · 2005. 7. 11. · erasure coding for...
TRANSCRIPT
![Page 1: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/1.jpg)
Networked Embedded Systems and the ZebraNet Project:
Experiences and Challenges
Prof. Margaret MartonosiDept. of Electrical EngineeringPrinceton University VET
TES EN
NOV TAM TVM
![Page 2: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/2.jpg)
Acknowledgments
Yong WangOther ZebraNet students: Ting Liu, Chris Sadler, Pei ZhangOther Princeton colleagues: Prof. Li-Shiuan Peh, Prof. Steve Lyon, Prof. Dan RubensteinOther folks: Sushant Jain, Kevin Fall
NSF ITR since 2002
![Page 3: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/3.jpg)
Challenges in Researching Networks of Embedded Sensors
Constrained optimization on highly-varying apps– Good system performance under energy, size, bandwidth
constraintsFast-changing hardware – Difficult to manage software changes, amortize software
development effortsInevitably-changing software (Learn->change->learn->change…)– Manage on-the-fly updates post-deployment
“New world” new metrics for success– Autonomy, reliability matter a lot– Need good infrastructure to evaluate…
![Page 4: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/4.jpg)
Our work: zebras + …
![Page 5: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/5.jpg)
ZebraNet as Biology ResearchGoal: Biologists want to track animals long-term, over long distances– Interactions within a species? – Interactions between species?– Impact of human development?
Current technology is limited:– VHF Triangulation is difficult & error-
prone– GPS trackers limit data to coarse
sampling and require collar retrievalOverall, energy and info retrieval are key limitersPeer-to-peer offers opportunity to improve
![Page 6: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/6.jpg)
ZebraNet as Computing Research
Research QuestionsProtocols and mobility? Energy-efficiency? Software layering design?
Data
Base station (car or plane)
Data
DataStore-and-forward communications
Data
Tracking node with CPU, FLASH, radio and GPS
ZebraNet vs. Other SensorNetsAll sensing nodes are mobileLarge area: 100’s-1000s sq. kilometers“Coarse-Grained” nodesGPS on-boardLong-running and autonomous
![Page 7: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/7.jpg)
Basic System Operation
TrackingNode A
AA TrackingNode B
BB
BB
AA
![Page 8: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/8.jpg)
Daily/weekly;Car or Plane
Basic System Operation
TrackingNode C
CC
TrackingNode B
BB AA
BB AA
CC
Potentially much laterand far from node A…
CC BB AA
![Page 9: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/9.jpg)
Abstraction Layers and Research Challenges
Application
Protocols & Data Collection
Middleware/OS
Hardware
Physical Design,Circuits, Antennas, …
Impact of app constraints and needs on design?– energy, weight, lifetime,
datarate– latencies, data rates,
accuracyManaging constant change of applications and hardwareInfo passing between interface layers?Portability across apps?
![Page 10: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/10.jpg)
MicrocontrollerTI MSP43016-bit RISC
~2KB RAM, 60KB ROM8MHz/32KHz dual clock
ATMEL FLASH~80 days data capacity
GPS10-20s position fix time
Maxstream Radio
0.5-1mile transmit range
Power supplies, solar modules, charging circuits
ZebraNet Hardware Design
312 mW8MHz +radio rcv
780 mW8MHz + radio xmit
568 mW8MHz w/ GPS
19.32 mW8MHz CPU
9.6 mW32Khz CPU
PowerMode
![Page 11: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/11.jpg)
January, 2004 ZebraNet Hardware
![Page 12: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/12.jpg)
January, 2004: Initial test deployment
Eat/sleep/work at MpalaResearch Centre near Nanyuki KenyaDeploy collars at Sweetwaters Preserve, also near Nanyuki Kenya
![Page 13: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/13.jpg)
First Deployment ResultsBiology
Zebras affected by collar first day (head shaking) but little thereafterFirst night-time zebra movement data: animals appear to explore more wooded areas and gullies at night
EngineeringRadio range: <1km in final collar packaging, – Disappointing vs. NJ tests– Answer: Choose cobbler well…
Communication protocols: generally worked as plans, although duplicated ACK packets improved their performance
![Page 14: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/14.jpg)
15 months later…
Collar on Zebra for roughly 1.25 years– Hot sun, tough weather.
Zebra attacked/killed by a lion (not unusual, probably not related to collar)How did the collar do?
![Page 15: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/15.jpg)
Next StepsSecond Deployment: June/July, 2005 (NOW!!)– Amorphous silicon solar cells with Tefzel
coating– Leather rather than butyl rubber collar– New version of msp processor– New radio– New GPS– New collar design– Improved code (latencies, interrupt
handling….)
![Page 16: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/16.jpg)
![Page 17: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/17.jpg)
![Page 18: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/18.jpg)
![Page 19: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/19.jpg)
![Page 20: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/20.jpg)
![Page 21: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/21.jpg)
Beyond zebras…
![Page 22: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/22.jpg)
Systems research feeding into mobile wireless application domains…
ZebraNet
CPU and HardwareDesign
Software Systems Research
Networking,Analytic Modeling,
Simulation
Smart Buildings Traffic Management
. . .
![Page 23: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/23.jpg)
Challenges: Another look…
Lack of stable application drivers on which to experiment– Applications very much on-the-move
right now. Hard to pin down general characteristics
Lack of stability in underlying hardware and software– What are the default hardware and
software platforms to assume?– How to support their ongoing change?
Other: Radio range, security, …Lack of good experimental infrastructure– Testbeds– Simulators
Likely to change
Look at these first!
Unlikely to change
Unlikely to change
Unlikely to change
![Page 24: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/24.jpg)
Changing Applications, changing platforms
Ongoing change is the mark of all dynamic/interesting areas of computingRather than wait for change to slow down, embrace it and look for ways to manage/encourage it
Abstraction layersDynamic and model-driven adaptation
![Page 25: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/25.jpg)
Abstraction Layers and the Impala OS
![Page 26: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/26.jpg)
Abstraction Layers and Research Questions
Application
Protocols & Data Collection
Middleware/OS
Hardware
Physical Design,Circuits, Antennas, …
Portability across apps?Impact of app constraints and needs on design?– energy, weight,
lifetime, datarate– latencies, data rates,
accuracyInfo passing between interface layers?
![Page 27: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/27.jpg)
Impala Middleware/OS Layers
Radio GPS FLASH TimerCPU WDT
Access and Control to All Devices
Application 1 Application 2 Application 3 Application 4
GPS Data EventRadio Packet EventTimer Event
Asynchronous Network TransmissionProtected FLASH AccessApplication Timer Control
GPS Data Event HandlerApplication Timer Event HandlerNetwork Packet Event HandlerNetwork Send Done Event HandlerSystem Clock Time Read
Adapter Updater NetworkSupport
OperationScheduler
EventFilter
![Page 28: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/28.jpg)
DB
Impala Code Updates
High Node MobilityConstrained BandwidthWide Range of Updates
Incomplete UpdatesUpdates vs. ExecutionOut of order Updates
ZebraNet Characteristics Design Implications
UpdaterUpdate
A C Node
On a single sensor node Full network
![Page 29: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/29.jpg)
Code and Data Size:Memory Footprint of Impala Layers
Code Size
0
10
20
30
40
50
60
70
KB
Firmware ImpalaApplication Unused
Data Size
0
500
1000
1500
2000
2500
Byte
s
Firmware ImpalaApplication Unused
![Page 30: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/30.jpg)
Dynamic & Model-Driven Adaptation
![Page 31: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/31.jpg)
Adjusting to uniquely mobile characteristics
Mobile networks have gained a lot by building off techniques from static wired networks.But they have lost a lot too– Their ability to do broadcast-for-free is rarely
leveraged as fully as it could be.– Recognizing disconnected or disrupted
optimization and customizing for it– Ability to adapt to changing network and exploit it
intelligently
![Page 32: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/32.jpg)
Model-based Mobility-Adaptive Protocols
Our Approach
Traditional Approach
![Page 33: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/33.jpg)
Example: Markov Model based on route cache lifetimes
Markovroute cache
model
22%2.12.78 pkts/s
27%34.14 pkts/s
16%5.26.22 pkts/s
% latency improvement
Markov approach
Original (DSR)
Traffic Rate
With Yong Wang, LS Peh, under submission
![Page 34: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/34.jpg)
Tolerating Sparse and High-Disruption Wireless Networks
What if a source->dest route never exists? Or rarely exists?Need to engineer more opportunistic approachesModel-based: if you can learn about mobility patterns, you can exploit this knowledge to know how to routeCover-your-bases: If not, use erasure-coding to reduce risk
![Page 35: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/35.jpg)
Erasure Coding for Disruption-Tolerant Wireless Networks
Basic Approach– For given “replication budget”, replicate via erasure-
coded packets, rather than basic redundancy. – Partial packet delivery can be used to reconstruct original
messageIntuition– Erasure-coding reduces “risk” due to outlier bad links– Erasure-coding based approach has the lowest worse-
case pkt delivery latency
With Yong Wang, Sushant Jain, Kevin Fall, Sigcomm WDTN 2005
![Page 36: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/36.jpg)
Packet delivery latency for Erasure-Coding and other approaches
Flooding has bestworst-case time
Erasure-coding:better worst-casethan other replication
![Page 37: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/37.jpg)
Beyond zebras: Final challenge is leveraging lessons/infrastructure over many apps…
ZebraNet
CPU and HardwareDesign
Software Systems Research
Networking,Analytic Modeling,
Simulation
Smart Buildings Traffic Management
![Page 38: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/38.jpg)
Summary and wishlist
Better built-in infrastructure for real-system wireless measurements– Testbeds, especially large-nodecount and unique
physical placementsIncreased availability of “real-world” mobility tracesContinued and increased research focus on:– composition of algorithms into real systems– and on how real-systems behaviors impact theoretically-
good algorithms– Portable hardware/software interfaces and clean models
for managing complex parallelism in these systems
![Page 39: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/39.jpg)
The Princeton ZebraNet Project: Mobile Sensor Networks for Wildlife Tracking
Grads: Pei Zhang, Chris Sadler, Ting Liu, IlyaFischoff, Yong Wang, Philo Juang.Profs: me, Dan Rubenstein, Steve Lyon,Li-Shiuan Peh, Vince Poor.Undergrads: Julie Buechner, Chido Enyinna, Brad Hill, Kinari Patel, Karen Tang, Jeremy WallDepartments of EE, CS, and Biology at PrincetonFunded by NSF ITR since 9/2002
ZebraNet Folks at Mpala Research Centre,near Nanyuki, Kenya. January 2004.
![Page 40: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/40.jpg)
More questions?
For more info, see papers: ASPLOS02, PPOPP03, Mobisys04,SenSys04
… and my webpagewww.princeton.edu/~mrm
![Page 41: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/41.jpg)
Preliminary results
![Page 42: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/42.jpg)
Why routing in disrupted network is hard (I)
Node mobility is very unpredictable and highly varying for many emerging applications
– Zebras move in phases with very distinct mobility characteristics– Data mule links may have very long delays – No model available for these emerging movement traces yet
• Interesting and challenging for the modeling and measurement community
• Hard to generalize– Statistics models alone may not be suitable since mobility shows very
different features in different states• From the system perspective, such model/measurement is also great
importance to in understanding the design space and drive new designs
![Page 43: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/43.jpg)
Why routing in disrupted network is hard (II)
Communications maybe asynchronous and contemporary due to intermittent connections
– Traditional routing is based on end-to-end connection– Store-and-forward is the most widely used forwarding algorithm in
current DTNs– Based on various replication schemes
_flooding _direct_simpleReplication _history/probabilistic
– Performance heavily dependent on the relays selected• Delay • Data delivery rate
![Page 44: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/44.jpg)
Why routing in disrupted network is hard (III)
Constrained device and varying quality links
– Nodes are energy constrained– Cannot afford expensive operations like period broadcast or
neighbor probing– Especially true when workload is high
– Wireless bandwidth is limited– Radio range is very limited in our testbed environment
– Strong interference and unpredictable quality– Radio transmission quality varies with environment, traffic
workload, to name a few– No golden rule that a link is always good– Shortest-path based on link hops may not be perform well
![Page 45: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/45.jpg)
Why routing in disrupted network is hard (IV)
Traffic is application-dependent and unpredictable
– When traffic workload increases, such as from a demand for high-sampling rate in wireless sensor networks, some of the underlying assumptions used by current routing may break
– One example: ETX routing• The method to measure link quality by snooping data packets or
sending probing packets fail.• Effective adaptation requires situation-aware so that mechanisms
that work in current scenario will be used.
![Page 46: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/46.jpg)
Why routing in disrupted network is hard (V)
Evaluation
– Lack of testbed and simulation tools– Partly due to the lack of models for such
emerging applications– No open testbed for widely use, like motelab or
mirage in static sensor network area• It will be cool to have a live znetlab for experiment
![Page 47: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/47.jpg)
Our Approach: Put intelligence into decision-making
Smart adaptation based on accurate understanding of mobility, traffic and link quality measurement, etc.
Ia. MARioIb. Model-based routingII. Erasure Coding based DTN RoutingIII/IV. MetricMap
![Page 48: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/48.jpg)
Mobility Adaptive Routing
Protocolswith hand-tuned
pre-programmed constants
Mobilityas a black-box
Traditional Approach
![Page 49: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/49.jpg)
optimizations
MARiowith mobility-adaptive strategies
(this is different from recently proposed mobility-controlling approach)
Route lifetime abstractions (RLA)
Our Approach
![Page 50: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/50.jpg)
• We use DSR as the base protocol and implement MARio atop it• The performance comparison is among the following protocols:
• Base DSR, Offline (GA), Online (LTA) and ORPD (Oracle)
• Performance breakdown demonstrates operation (prefetch/decay) accuracy:
• Prefetch only, Decay only, Prefetch and decay
Raw performance Performance breakdown
With Yong Wang, LS Peh, Mobihoc poster 2004, mc2r 2005
![Page 51: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/51.jpg)
ZebraNet Hardware: Time-Lapse View…Aug ‘02
Jan ‘03
Aug ‘03 Oct ‘03
![Page 52: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/52.jpg)
Impala: Middleware/OS Support for Application/Protocol Modularity
Monolithic Approach:Source code hardwired to particular situationsApp responsible for adapting situation/version choicesDifficult to debug, maintainDifficult to update on the flyDifficult for other apps to reuse
Layered Approach:OS provides network and event-handling servicesMiddleware adapts, updates apps, protocols dynamicallyNew protocols can be plugged in at any time
CBBC
A D
Impala Layer
A B
D
A B
D
Individual Protocols
Aggregate Protocol
![Page 53: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/53.jpg)
Forwarding Algorithms Comparison
r highest ranked
New contactAll nodesHistory(r)
kr (k>=1)first contacts (k is
related to coding algorithm)
New contactSource onlyErasure Coding (ec-r)
r first contactsNew contactSource onlySimple replication(r)
Destination only
DestinationSource onlyDirect
All newNew contactAll nodesflood
To whomWhenWhoAlgorithm
![Page 54: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/54.jpg)
Analyzing node contact times from zebra mobility trace
Duration of deployment
Inter-contact timeContact duration
![Page 55: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/55.jpg)
MetricMap: a machine learning approach to estimate link quality
Discover correlations between local metrics (feature vector) and link quality indicator (goal)
Buffer size, channel clear access, depth, RSSI, etc.Hardware-provided link quality indicator
Closely related to received signal strength indicatorAlso reflects other factors
Derive inference rules using standard machine learning algorithm
Weka packageC4.5 (decision tree) and JRip (decision rules)training sets collected by running on a Berkeley-MicaZ
based real-world sensor network testbed
![Page 56: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/56.jpg)
Deployed on MoteLab and Mirage
With Y. Wang, L.S Peh, under preparation
![Page 57: Networked Embedded Systems and the ZebraNet Project: … · 2005. 7. 11. · Erasure Coding for Disruption-Tolerant Wireless Networks Basic Approach – For given “replication budget”,](https://reader033.vdocuments.net/reader033/viewer/2022052018/60314c99b56bfb2416015532/html5/thumbnails/57.jpg)
Preliminary Results
MetricMap improves on previous approaches:– 10-40% in terms of packet latency– ~80% in terms of packet success rate– Also improves fairness properties
Benefits especially pronounced at high offered loads.– MetricMap is good at distinguishing packet drops
due to link quality from packet drops due to congestion