Wireless Sensor Networks: Time for RealTime

1
Wireless Sensor NetworksTime for Real-Time?
Professor Jack Stankovic Department of Computer
Science University of Virginia February 2009
2
Ad Hoc WSN

• Sensors
• Actuators
• CPUs/Memory
• Wireless Radio
• Power Limited

Self-Organizing
3
Mica2, Mica2Dot, MicaZ

• ATMega 128L 8-bit, 8MHz, 4KB EEPROM, 4KB RAM,
  128KB flash.
• Chipcon CC100 or CC2420.

Mica2
Mica2Dot
4
Outline

• Motivation do WSN have RT constraints?
• Can we apply real-time technology?
• If so, what technology?
• Should we apply real-time technology?
• Real-Time Scheduling and WSN
• Real-Time Control and WSN
• Final Thoughts

5
Computing in Physical Systems
Road and Street Networks
Environmental Networks
Industrial Networks
Open, Heterogeneous Wireless Networks
with Sensors and Actuators
Battlefield Networks
Building Networks
Vehicle Networks
Body Networks
6
Real-Time Constraints?

• Many types
• Hard deadlines
• Hard deadlines and safety critical
• Soft deadlines
• Time based QoS

7
Example Group Management (Tracking)
Base Station
8
Deadlines

• If we have enough late messages within groups we
  can lose the track
• Not straightforward deadline
• Tied to redundancy, speed of target
• If messages dont make it to base station in hard
  deadline we miss activating IR camera
• If we dont act by Deadline D truck carrying bomb
  explodes safety critical

9
What RT Technology?
10
Real-Time Systems

• A real-time system is one in which the
  correctness of the system depends not only on the
• logical result of computations, but also
• on the time at which the results are produced.

11
Real-Time Constraints

• Hard and Soft RT
• Safety critical

V
V
D
D
V
D
minus infinity
12
Technology (pointers)

• Rate Monotonic Analysis
• A practitioners handbook for real-time analysis
  Klein, et. al.
• Earliest Deadline First (EDF) theory
• Deadline scheduling for real-time systems
  Stankovic et. al.
• Control Theory based scheduling
• Deterministic OS kernels
• Real-Time Operating Systems Stankovic et. al.
• Predictability and (Hard) Guarantees
• Hard Real-Time Computing Systems - Buttazzo

13
Can We Apply RT Technology

• Basis of that technology
• Deterministic underlying models
• Worst case assumptions
• Highly controlled (closed) environments

14
WSN Uncertainties

• Highly dynamic wireless communication
• Arbitrary Interference
• Environment
• Parallel communications
• Other wireless systems

Highly Controlled Highly
Uncontrolled
Classical RT WSN
15
Communication
Irregular Range of A
A and B are asymmetric
16
Spatial Impact

RSSI measured in a parking lot
17
Temporal Impact

RSSI measured every hour
18
Adaptive Transmission Power Control (ATPC)
See our paper in SenSys
19
Uncertainties -Voids
Left Hand Rule

Destination
VOID
Source
20
Other Uncertainties

• Energy depletion
• Nodes fail
• Environmental disturbances
• Other co-located systems (mobile and not)

RT Impossible?
21
Any Good News?

• Sensors gather redundant information
• Virtualization
• Create reliable system out of unreliable
  parts
• Can we do this in time
• Can we tame the uncertainties

22
Guarantees

• Function of Assumptions
• Classical Hard Real-Time
• Closed/controlled system
• Careful set of assumptions
• Gap between assumptions and system is often small
• Subject to max number of losses/errors

• Function of Assumptions
• WSN
• Open system
• Careful set of assumptions
• Gap may be large
• Subject to max number of losses/errors

23
Real-Time Technology

• Velocity Monotonic
• Exact Characterization
• Deadline Partitioning
• SW-based Control Theory

24
Sensor Net Routing

• End-to-end
• Real-time
• Collisions
• Congestion
• Power
• Security
• Mobility

Destination
Source
Base Station
Assumption Nodes know location
25
Velocity Monotonic Scheduling (VMS)

• Consider Time and Distance
• Loc.(Dest.) Loc.(Source) Distance
• Distance/Deadline Velocity

See our RAP paper
26
VMS

• Static VMS (compute at source)

• Fixed velocity on each hop
• Vdis(x0,y0,xd,yd)/D

• Dynamic VMS

• Adapt velocity at intermediate node
• Vi dis(xi,yi,xd,yd)/Si
• Slack Si D - elapsedTime

F(remaining distance)
27
Example
D
dis 90 m D 2 s V 45 m/s HIGH Priority
A
C
B
dis 60 m D 2 s V 30 m/s LOW Priority
28
Deadline Miss Ratio
Deadline Miss Ratio FCFSgtDSgtDVM,SVM
Overall Deadline Miss Ratios with deadlines (5,10)
29
Best Effort

• Runtime algorithm
• Can we develop a VMA (velocity monotonic
  analysis) analogous to RMA?
• More rigorous than best effort

30
Next Example

• Static off-line
• Early assessment
• Careful set of assumptions

31
WSN Schedulability Analysis

• Given a sensor network and multiple periodic
  real-time streams, can all streams meet their
  time constraints in the network?
• Carefully state assumptions
• Result produce a feasible communication link
  schedule

32
Exact Characterization

• Given
• Fixed Topology of the WSN
• Communication Matrix
• Interference Matrix
• Set of Streams
• D, P, Start_Time, Transmission_Time, Source and
  Destination
• Fixed Routing
• Clock sync slots
• Is there a schedule for all streams to meet all
  constraints?

33
Schedulability Analysis
Stream specification
Network topology
Communication parameters
Result
schedulable Communication Link from
node 1 to 2 is assigned to stream 1 at time slot
1
Communication Link from
node 3 to 5 is assigned to stream 3 at time slot
1
34
Exact Characterization - Algorithm

• 1. Order streams by importance (vel.)
• 2. Find and allocate free links to most important
  stream
• 3. Update remaining streams conditions based on
  new allocation
• 4. If any streams still pending, then go to step
  1.

35
Example per stream
36
Workloads

• Known workloads
• All, most common, expected, outlier, key
  workloads
• Design time
• Provides an understanding of capabilities of the
  system
• Guarantees subject to assumptions
• Allocate per stream or per link
• If fails, know that up to Stream k the system is
  OK
• Redesign

37
Robustness

• Change topology
• Change routing
• Communication and Interference matrices as a
  function of time
• Allow for message faults (retransmissions)

38
Analogous to the cylinder packing problem
39
More Rigorous than VMS

• Exact analysis (uses velocity)
• Many assumptions
• Off-line (design time) estimation
• Robustness studies possible

40
Case Study

• System-wide view of real-time
• Surveillance and tracking application
• Apply classical real-time approach
• Deadline partitioning

41
VigilNet
1. An unmanned plane (UAV) deploys motes
Zzz...
Sentry
2. Motes establish a sensor network with power
management
3. Sensor network detects vehicles and wakes up
the sensor nodes
42
VigilNet Architecture
43
Deadline Partition

• Reduce the design space through deadline
  partition.
• Assign sub-deadlines for each stage in the
  system to achieve the overall deadline
• Example to follow
• 6 stages
• Overall deadline 5 seconds

44
Stage 1. Target Triggers the First Sentry

• Initial Activation (Tinitial)
• Nodes are either sentry/non-sentry
• For Sentry node periodic sleep (P)

Tinitial the time until the a sentry detects
the target
45
Stage 2 Sentry Ensures it is a Real Target

• Initial Activation (Tinitial)
• Initial Target Detection (Tdetect)
• Hardware delay response
• Sampling delay, accumulate sampling
• Running the detection algorithm

Tdetection the time until the first sentry
confirms detection
46
Stage 3 Wake Up Neighboring Nodes

• Initial Activation (Tinitial)
• Initial Target Detection (Tdetect)
• Wakeup (Twakeup)
• Time to from a group around the target.

Twakeup the time to broadcast wakeup messages
47
Stage 4 Sentry Aggregates the Reports

• Initial Activation (Tinitial)
• Initial Target Detection (Tdetect)
• Wakeup (Twakeup)
• Group Aggregation (Taggregation)
• Establish a group to track the target and define
  a logical entity
• Wait for multiple messages from group members to
  validate detection

Taggregation the time to aggregate messages
48
Stage 5 Send Aggregated Results to Base Station

• Initial Activation (Tinitial)
• Initial Target Detection (Tdetect)
• Wakeup (Twakeup)
• Group Aggregation (Taggregation)
• End-to-end delay (Te2e)
• Relay message to base-station

Te2e the time relay the message to base-station
49
Stage 6Base Station Processing

• Initial Activation (Tinitial)
• Initial Target Detection (Tdetect)
• Wakeup (Twakeup)
• Group Aggregation (Taggregation)
• End-to-end Delay (Te2e)
• Base Processing (Tbase)

Tbase the time to do summary at the Base station
50
System Evaluation
51
200 XSM Motes
52
Delays
Average detection delay 2.42 seconds Average
classification delay 3.56 seconds Average delay
to get velocity estimation 3.75 seconds
53
Control Theory
Controlled variable
Overshoot
Steady state error
??
Reference
Steady State
Transient State
Time
Settling time
54
Sensor Net Routing

• End-to-end
• Real-time
• Collisions
• Congestion
• Power
• Security
• Mobility

Destination
Source
Base Station
Assumption Nodes know location
55
SPEED
USE VELOCITY
See our Speed paper
56
SPEED Architecture
57
Evaluation
(Added Congestion)
E2E Delay Energy
Consumption
58
Performance
Figure A. E2E delay profile of DSR
Figure B. E2E delay profile of AODV
59
Performance
Figure. D E2E delay profile of SPEED
Figure C. E2E delay profile of GF
60
Final Thoughts

• We can and should apply RT Technology to WSN
• What is a guarantee
• Function of many different underlying assumptions
• But so is classical real-time scheduling theory
• What are the limits/usefulness of rigorous
  real-time analysis for WSN?