Dependable%20messaging%20in%20wireless%20sensor%20networks - PowerPoint PPT Presentation

View by Category
About This Presentation
Title:

Dependable%20messaging%20in%20wireless%20sensor%20networks

Description:

Dependable messaging in wireless sensor networks – PowerPoint PPT presentation

Number of Views:32
Avg rating:3.0/5.0
Slides: 51
Provided by: csWa9
Learn more at: http://www.cs.wayne.edu
Category:

less

Write a Comment
User Comments (0)
Transcript and Presenter's Notes

Title: Dependable%20messaging%20in%20wireless%20sensor%20networks


1
Dependable messaging in wireless sensor networks
  • Hongwei Zhang
  • http//www.cs.wayne.edu/hzhang

2
(No Transcript)
3
Wireless sensor networksinnovative ways of
interacting with physical world
  • Science ecology, seismology, oceanography

Engineering structural monitoring, factory
automation, precision
agriculture
Daily life traffic control, health care,
home security, virtual tour
4
Sensor nodes
  • A XSM sensor node (2004)
  • 8MHz CPU, 4KB RAM, 128KB ROM
  • Chipcon CC1000 radio 19.2 kbps
  • Infrared, acoustic, and magnetic sensors
  • Sounder
  • ?
  • Many more (2001 - )

5
Sensor networks
Redwood ecophysiology
New applications, startup companies keep
emerging
Wind response of Golden Gate Bridge
Intruder detection, classification, and tracking
6
Are sensornets easy to use today?
  • Each project may well take
  • 5 professors
  • 10 Ph.D.
  • 20 Master and undergraduate
  • 40 labors
  • A few months/years of hard work

7
Challenging aspects of sensor networks
  • Dynamic, unreliable, and interference-prone
    wireless channels
  • Reliable messaging
  • Indoor testbed at OSU 3 feet node separation
  • 300 data points for each distance, with each data
    points representing the status of 100 broadcast
    transmissions

8
Challenging aspects of sensor networks
  • Dynamic, unreliable, and interference-prone
    wireless channels
  • Reliable messaging
  • Resource constraints (e.g., bandwidth, energy,
    memory)
  • Resource-efficient services, sensornet
    architecture

9
Challenging aspects of sensor networks
  • Dynamic, unreliable, and interference-prone
    wireless channels
  • Reliable messaging
  • Resource constraints (e.g., bandwidth, energy,
    memory)
  • Resource-efficient services, sensornet
    architecture
  • Application diversity (e.g., traffic patterns,
    QoS requirements)
  • Application-adaptivity

10
Challenging aspects of sensor networks
  • Dynamic, unreliable, and interference-prone
    wireless channels
  • Reliable messaging
  • Resource constraints (e.g., bandwidth, energy,
    memory)
  • Resource-efficient services, sensornet
    architecture
  • Application diversity (e.g., traffic patterns,
    QoS requirements)
  • Application-adaptivity
  • Complex faults and large system scale
  • Dependability despite fault complexity and system
    scale

Node/link failure, state corruption, system
signal loss, malfunctioning sensor, etc
Increased overall probability of fault occurrence
Fault propagation
11
Challenging aspects of sensor networks
  • Dynamic, unreliable, and interference-prone
    wireless channels
  • Reliable messaging
  • Resource constraints (e.g., bandwidth, energy,
    memory)
  • Resource-efficient services, sensornet
    architecture
  • Application diversity (e.g., traffic patterns,
    QoS requirements)
  • Application-adaptivity
  • Complex faults and large system scale
  • Dependability despite fault complexity and system
    scale
  • Heterogeneity
  • Architecture and service provisioning in
    integrated systems

12
Challenging aspects of sensor networks
  • Dynamic and potentially unreliable wireless
    channels
  • Reliable messaging
  • Resource constraints (e.g., bandwidth, energy,
    memory)
  • Resource-efficient services
  • Application diversity (e.g., traffic patterns,
    QoS requirements)
  • Application-adaptivity
  • Complex faults and large scale
  • Dependability irrespective of scale
  • Growing heterogeneity
  • Architecture and service provisioning in
    integrated systems

13
Our position
  • Reliability and efficiency via in-situ estimation
    and adaptation
  • Sensornet messaging architecture book chapter
  • Data-driven link estimation and routing
    INFOCOM06, WiNMee06, RTCSA05
  • Packing-oriented scheduling US Patent pending
  • Reliable and real-time data transport
    MobiHoc05, COMCOM, COMNET, JACIC
  • Scalable dependability via local stabilization
  • Locally-stabilizing shortest path routing
    PODC02, DSN03, PODC05, ToN, COMNET

14
Our systems deliverables
  • Provided dependable messaging software for both
    the mote and 802.11 networks of our sensornet
    systems (DARPA NEST)
  • A Line in the Sand COMNET
  • 100 MICA2s (TinyOS) 10 X 20 meter2 field
  • ExScal RTCSA05, RTSS05, DCOSS05, MobiSys05
  • 1,000 XSM, 200 MICA2s, 200 802.11 Stargates
    (Linux)
  • 1,300 X 300 meter2 field
  • Provided Maté/VM support for application-adaptive
    messaging
  • Motorola Labs has filed this work for a US patent
  • Co-designed the wireless networks of Kansei, a
    sensornet testbed for at-scale experiments
    IPSN/SPOTS06
  • My messaging software are also adopted in
  • the 10,000 nodes sensornet system being deployed
    by Northrop Grumman Corporation
  • Kestrel Institute

15
Our position
  • Reliability and efficiency via in-situ estimation
    adaptation
  • Sensornet messaging architecture
  • Data-driven link estimation and routing
  • Packing-oriented scheduling
  • Reliable and real-time data transport
  • Scalable dependability via local stabilization
  • Locally-stabilizing shortest path routing

16
Outline
  • Sensornet messaging architecture
  • Data-driven link estimation and routing
  • Packing-oriented scheduling
  • Reliable and real-time data transport

17
Motivation
  • The monolithic approach in sensornet systems
    engineering is inefficient
  • It is desirable to identify the common services
    used in different applications
  • Sensornet messaging demands services not found in
    existing networks
  • Deal with dynamic wireless links which are
    subject to the impact of both environment and
    application traffic pattern
  • Support in-network processing
  • Convergecast as a typical communication pattern

18
SMA sensornet messaging architecture
  • Supporting application-specific in-
  • network processing and QoS
  • requirements
  • Explicit input from application

Application
Traffic-adaptive link estimation and routing (TLR)
  • Forming the basic communication
  • structure in a traffic-adaptive
  • manner
  • No explicit input from application

H. Zhang, A. Arora, P. Sinha, L. J. Rittle,
Messaging in Sensor Networks Addressing Wireless
Communications and Application diversity,
Handbook of Real-time and Embedded Systems, CRC
Press, to appear
19
Traffic-adaptive link estimation and routing
  • Network conditions vary significantly across
    different interference scenarios (e.g., up to
    39.26) variations change with distance

20
Application-adaptive structuring
0
2
1
3
4
5
21
Application-adaptive scheduling
0
Hold wait
2
1
3
4
5
22
Outline
  • Sensornet messaging architecture
  • Data-driven link estimation and routing
  • Packing-oriented scheduling
  • Reliable and real-time data transport

23
Link estimation and routing
ExScal
  • State of the art beacon-based link estimation
  • Neighbors periodically exchange broadcast beacons
  • Estimate unicast properties via those of
    broadcast
  • Note application data is transmitted using
    unicast
  • Drawbacks of beacon-based estimation
  • Network condition experienced by beacons may not
    apply to data
  • Traffic pattern may change quickly (especially in
    event-detection applications), and Traffic
    pattern affects link properties due to
    interference
  • It is hard to precisely estimate unicast
    properties via those of broadcast beacons
  • Temporal link properties (e.g., correlation,
    variance) non-trivial to model, and not
    considered in well-known approaches such as ETX

24
Impact of temporal properties
(actual unicast reliability estimated unicast
reliability)
  • There is significant estimation error, especially
    in the transitional region error changes with
    distance and interference pattern

25
Implications
  • Beacon-based link estimation is not
    traffic-adaptive and tends to be imprecise (which
    degenerates network performance)
  • Should we explicitly model the impact of traffic
    patterns and temporal link properties?
  • Our answer data-driven link estimation and
    routing
  • Estimate unicast link properties via data
    transmission itself
  • Circumvent the complexity and drawbacks of
    beacon-based estimation

Non-trivial
Provided reliable and efficient routing in
ExScal, etc. H. Zhang, A. Arora, P. Sinha, Learn
on the Fly Data-driven Link Estimation and
Routing in Sensor Network Backbones, INFOCOM 2006
26
A challenge for data-driven routing
  • Uneven link sampling a link will be sampled only
    if it is used as the forwarder

?
?
?
Ri
?
A
R
t0
R is better
Ri becomes better
Better routes may not be used.
27
Countermeasure exploratory sampling
  • Proactively sample alternative neighbors, by
    using them as forwarders
  • Control the frequency to reflect the goodness of
    the current forwarder
  • Sample a neighbor with the probability of it
    being the best forwarder

t0
Ri
time
A
R
28
End-to-end MAC latency
  • Compared with ETX and PRD, LOF reduces MAC
    latency by a factor of 3
  • LOF has the smallest MAC latency compared with
    L-, showing the importance of
  • proper exploratory sampling
  • not assuming geographic uniformity

75-percentile
median
25-percentile
29
Average number of unicast transmissionsper
packet received
  • Compared with ETX and PRD, LOF improves
    efficiency by a factor of 1.49 and 2.37
    respectively
  • LOF is more efficient than L-

30
Links used reliability and length
of transmission failures
average link length
  • LOF uses reliable links
  • 1112 and 786 failures in ETX and PRD
    respectively only 5 failures in LOF
  • L-ns uses reliable but shorter links than LOF does

31
Outline
  • Sensornet messaging architecture
  • Data-driven link estimation and routing
  • Packing-oriented scheduling
  • Reliable and real-time data transport

32
Packet packing an example of in-network
processing
  • Aggregate short packets into longer ones
  • short information unit (lt 10 bytes), yet long
    packet header (e.g., up to 31 bytes for 802.15.4)
    and other overhead
  • Allowable payload length is long (e.g., up to 102
    bytes for 802.15.4)
  • Packets do not always need to be delivered
    immediately
  • Benefits
  • Reduced cost (e.g., header overhead, energy spent
    in radio wakeup) in transmitting each information
    unit
  • Reduced channel contention

33
Packing-oriented scheduling
  • Objective
  • Schedule packet transmissions to improve the
    degree of in-network packing, while satisfying
    application requirements on the timeliness of
    packet delivery
  • Approach utility based scheduling, where utility
    is defined as the expected reduction in
    (amortized) cost of packet transmission

Immediately transmit the packet if Up gt Ul
US Patent pending Provided application-adaptive
messaging in the sensornet systems of Motorola
Labs, etc.
34
Packing ratio
  • Compared with noPacking and simplePacking,
    intelliPacking improves the packing ratio by a
    factor of 5.25 and 3.5 respectively

35
Energy efficiency
Number of transmission per information unit
received
Number of receptions per information unit received
  • Compared with noPacking and simplePacking,
    intelliPacking improves energy efficiency by a
    factor of 3.22 and 1.85 respectively

36
Information delivery reliability
  • Compared with noPacking and simplePacking,
    intelliPacking improves information delivery
    reliability by a factor of 12.92 and 12.77
    respectively

37
Outline
  • Sensornet messaging architecture
  • Data-driven link estimation and routing
  • Packing-oriented scheduling
  • Reliable and real-time data transport

38
Synchronous explicit ack (SEA)
A Line in the Sand
  • Retransmission does not help much, and may even
    decrease reliability and throughput
  • Similar observations when adjusting contention
    window of B-MAC and using S-MAC

Metrics RT 0 RT 1 RT 2
Reliability () 51.05 54.74 54.63
Latency (sec) 0.21 0.25 0.26
throughput (pkt/sec) 4.01 4.05 3.63
39
Stop-and-wait implicit ack (SWIA)
  • Again, retransmission does not help
  • Compared with SEA, longer latency and lower
    goodput/reliability
  • longer retransmission timer (coupled with
    blocking flow control)
  • more ACK losses, and thus more unnecessary
    retransmissions

Metrics RT 0 RT 1 RT 2
Reliability () 43.09 31.76 46.5
Latency (sec) 0.35 8.81 18.77
Goodput (pkt/sec) 3.48 2.58 1.41
40
Problem statement
  • How to achieve
  • close to 100 reliability?
  • close to optimal event throughput (implies
    real-time packet delivery)?
  • Elements of the solution RBC
  • Differentiated contention control
  • Window-less block acknowledgment
  • Fine-grain timer management

Provided reliable and real-time data transport in
sensornet systems A Line in the Sand, ExScal,
etc H. Zhang, A. Arora, Y.R. Choi, M. Gouda,
Reliable Bursty Convergecast in Wireless Sensor
Networks, MobiHoc 2005, Computer Communications
(Elsevier)
41
Experimental results for RBC
Metrics RT 0 RT 1 RT 2
Reliability () 56.21 83.16 95.26
Latency (sec) 0.21 1.18 1.72
Goodput (pkt/sec) 4.28 5.72 6.37
  • Retransmission helps improve reliability and
    goodput
  • close to optimal goodput (6.37 vs. 6.66)
  • Compared with SWIA, latency is significantly
    reduced
  • 1.72 vs. 18.77 seconds

42
Distribution of packet generation and reception
  • RBC
  • Packet reception smoothes out and almost matches
    packet generation
  • SEA
  • Quick packet reception
  • Many packets are lost
  • SWIA
  • Significant latency and packet loss

43
Summary
  • Architectural and algorithmic issues in
    dependable sensornet messaging
  • dealing with link dynamics and varying
    application properties, as well as their
    interactions
  • supporting in-network processing
  • Both the protocols and implementations have been
    deployed and verified in real-world systems

44
Ongoing and future research
  • Messaging architecture
  • Enrich and study SMA in typical sensornet
    applications
  • Identify typical messaging patterns (e.g.,
    convergecast, broadcast, and anycast) and study
    their impact on messaging architecture
  • Messaging protocols
  • Model wireless links and system dynamics, and
    study system performance limits
  • Provide predictable QoS to applications in the
    presence of system dynamics
  • Sensornet query and storage
  • Integration with existing pervasive systems and
    the Internet

45
Would like to learn more?
  • My CS homepage http//www.cs.wayne.edu/hzhang/
  • Courses
  • CSC 6290 (Fall 2006) fundamentals of computer
    networks
  • CSC 7290 (Winter 2007) advanced topics in
    networked systems
  • Research meetings
  • Contact information
  • Office 454 State Hall
  • Email hzhang_at_cs.wayne.edu
  • Tel 313-577-0731

46
Backup slides
  • Enabling technology in science (Culler05)

47
Three-age system of prehistoric societies
  • Stone age 2 millions years ago 6000 BC

Bronze age 3500 - 1200 BC
Iron age 1200 - 550 BC
48
Enabling technology for science
the complex
  • macroscope
  • P. Anthony 1969
  • J. de Rosnay, 1979

Perceive
the imperceptible
the atomic
the small
the far
49
Internet revolutionized information flow within
human community
Video conference
WWW
E-mail
Internet phones
IP picture frame
50
Breakdown of RBC
Metrics Metrics RT 0 RT 1 RT 2
Reliability () RBC 56.21 83.16 95.26
Reliability () RBC-NoDiffCtrl 54.90 77.19 82.29
Latency (sec) RBC 0.21 1.18 1.72
Latency (sec) RBC-NoDiffCtrl 0.22 1.12 1.52
Goodput (pkt/sec) RBC 4.28 5.72 6.37
Goodput (pkt/sec) RBC-NoDiffCtrl 4.04 4.13 4.12
  • RBC-NoDiffCtrl RBC without
    Differentiated Contention Control
  • Contention control plays an increasingly
    important role as RT (thus channel contention)
    increases
About PowerShow.com