Joint Mobility and Routing for Lifetime Elongation in Wireless Sensor Networks †

Download Report

Transcript Joint Mobility and Routing for Lifetime Elongation in Wireless Sensor Networks †

Joint Mobility and Routing for Lifetime
Elongation in Wireless Sensor Networks †
Jun Luo,
Jean-Pierre Hubaux
Laboratory of Computer Communications and Applications (LCA)
School of Computer and Communication Sciences
Ecole Polytechnique Fédérale de Lausanne (EPFL),
Switzerland
†This work was supported (in part) by National Competence Center in Research on Mobile Information and Communication
Systems (NCCR-MICS), a center supported by the Swiss National Science Foundation under grant number 5005-67322.
http://www.terminodes.org
1
Outline
• Motivations
– The longevity of sensor networks is important
– Traditional solutions to improving network lifetime
– A new trend that we follow
• Our approach: Joint mobility and routing
–
–
–
–
–
Basic idea
To move or not to move
Optimum mobility strategy
Better routing strategy
Implementation issues
• Simulation results
2
Longevity is Important
Our view of sensor networks: environmental
monitoring
Longevity is very important for many reasons:
deployment costs, environmental disturbance, ...
3
Traditional Solutions
• Basic principle: flow scheduling to balance the load
among forwarding nodes
• Example – Chang & Tassiulas [1]: linear programming
to maximize the time when the first node dies
• Problem: only the load among nodes that are at the
same distance from the base station is balanced.
• Consequence: the nearer a node is from the base
station, the higher the load it takes
4
A New Trend – Mobile Base Station
• Basic principle: picking up data from nodes with a
mobile base station (a mobile relay approach)
• Examples:
– Shah et al. [15]: Data MULE:
unpredictable mobility
– Chakrabarti et al. [16]: Predictable observer mobility
– Kansal et al. [26]: Controllable mobility
• Problem: the latency of data delivery is large.
• Consequence: these approaches are limited to certain
applications that do not have a stringent latency
requirement
5
Outline
• Motivations
– The longevity of sensor networks is important
– Traditional solutions to improving network lifetime
– A new trend that we follow
• Our approach: Joint mobility and routing
–
–
–
–
–
Basic idea and model
To move or not to move
Optimum mobility strategy
Better routing strategy
Implementation issues
• Simulation results
6
Basic Idea
• Move the base station to distribute over time
the role of “hot spots” (i.e., the nodes
around the base station) – a complement to
the traditional flow scheduling solution
• The data collection continues wherever the
base station is, so the solution does not
sacrifice latency – in opposition to the
mobile relay approach
7
Network Model
R

O

• A set of N nodes of Poisson
distribution with intensity 
within a circle for radius R
• Constant data rate 
between a node and a base
station
• An overall energy
consumption  of to receive
and transmit a unit of data
Base station
Sensor node
• Fixed transmission and
sensing range r
• Load-balanced routing
8
Problem Definition
• Network Lifetime : the time span from the sensor
deployment to the first loss of coverage.
• We convert the problem of maximizing network
lifetime to a min-max load problem:
Minimize
load N  max load n (strategies)
nN
because the area with the highest average load will most likely
lead to the first lose of coverage (which indicates the end of
the lifetime).
• Existing solutions to this problem involve only strategies
concerned with nodes (e.g., energy conserving routing).
• We intend to consider (base station) mobility
strategy and routing strategy together.
9
Modeling the Load of Sensor Nodes – I
We take w = r
We model average
load rather than
exact load, i.e.,
( S1  S 2 )
l oad n 
S2
(a) is based on Ganjali & Keshavarzian [23]:
• Rectangular envelope of width 2w for routing paths
• Two conditions for a node n to be on the way from x to B
(b) is a new model for nodes within the transmission range of
the base station
10
Modeling the Load of Sensor Nodes – II
We use S3 to
represent the
average load taken
by n. The model is
equivalent to the
previous one if:
( S1  S 2 )
 S3 
S2
• Computing the angle  for an arbitrary node is not trivial: it
cannot be achieved analytically.
• Fortunately, computing  for a node at the center is doable. So
we can use this value as an estimation for an arbitrary node
11
To Move or Not to Move – Static Case
load n 
( S1  S 2 ) 
S2 
 0.5 ( R 2  d 2 )
  d  r
 
2
  2 0.5r
 R 
d r
 r 2
Conclusion: the nodes
around the base
station use up their
energy much faster
than other nodes.
Therefore, their
lifetime upper bounds
the network lifetime.
12
To Move or Not to Move – Mobile Case
load n  
2
0
( R  d )  
1 R d

d


2
4R
2 R
2
2 2
2
2
2

  

Conclusion: mobile base station (even
with an arbitrary moving trajectory) does
help to balance the load. Further
improvements consist in:
• Reducing for  the hot spot (the center)
• Reducing the network size R
13
Optimum Mobility Strategy – I
Searching the trajectory space is not trivial, but the following
steps can reduce the space size:
• By defining periodic mobility as recurrent movements with a
constant period, we can consider aperiodic trajectory as
periodic mobility whose period is the same as the network
lifetime.
• CLAIM 2: Symmetric trajectory (rotation symmetry about
the center for all degrees) is at least as good as its nonsymmetric version.
Finally, we show that, by analytically comparison among all
symmetric trajectories,
CLAIM 3: The best trajectory is the network periphery (which
14
minimizes  ).
Optimum Mobility Strategy – II
CLAIM 2: Symmetric trajectory (rotation symmetry about the
center for all degrees) is at least as good as its non-symmetric
version.
M
load n  M 1 k 0 load n
where M  2 
PROOF:
k
 0
Conclusion:
We only need to
consider:
(i) movements on
concentric circles
(ii) identical
frequency
movements in
annuli.
1 2
1 2
 load n 
load n d 
load N d  load N


0
0
2
2
load N is the network load achieved by Τ 0
15
Optimum Mobility Strategy – III
CLAIM 3: The best trajectory is the network periphery (which
minimizes  ).
16
Better Routing Strategy
Conclusion: the scheme
does further improve the
network lifetime (see
simulations for details), but
analytical predication is
hard to achieve due to the
complicated situation
around the base station
trajectory.
The ideas:
• Although reducing R is impossible, it is possible to reduce
the radius of the section where short path routing is applied
• We divide the network into two sections and exploit the
redundant energy capacity of one section to compensate the
17
other one
Implementation Issues
• How to move? Robot + Node, see Butler [13] and Kansal et
al. [26] for details
• How to build routing path?
• Pre-computing can be done with a discrete movement
that coincides with sensor locations
• Periodical querying or routing information exchange
builds routing path automatically
• What about round routing? Trajectory based forwarding
(Niculescu & Nath [31])
• What if a non-circular network? Periphery mobility can be
nearly optimum, and it has a practical significance. A joint
strategy depends on the shape of the network region
18
Outline
• Motivations
– The longevity of sensor networks is important
– Traditional solutions to improving network lifetime
– A new trend that we follow
• Our approach: Joint mobility and routing
–
–
–
–
–
Basic idea and model
To move or not to move
Optimum mobility strategy
Better routing strategy
Implementation issues
• Simulation results
19
Simulation Setting
• High level simulator that ignores the MAC effects
• About 800 nodes deployed within a circle of 10 unites
with density  = 8/π
• Transmission range r = 1 unit
• Discrete movement (of the base station) consists of
several steps
• Emulating load-balanced routing by shuffling links
weights before searching for a routing path with
Dijkstra’s algorithm
20
Static vs. Mobile – II
21
Static vs. Mobile – II
Three reasons for the spikes
• Irregular topology
• discrete movement
• Imperfect emulation of load-balanced routing
22
Optimum Mobility
23
Better Routing
24
Conclusions
• Our contribution: using mobile base station to
extend the network lifetime
– Analytical models to characterize the energy consumption
patterns corresponding to certain movement strategies
– Opitmality results on the movement strategy
– Better routing strategy (than short-path routing)
• We also perform high level simulations to evaluate
the validity of our analysis.
• Future work:
– Implementations and field tests
– Detail model taking topology into account
– Mobile base station helps …
25
Let’s take a short break 
26
Load Balancing through Flow Scheduling
The ideas: All possible routes should be exploited
How? Off-line scheduling
27
A Linear Program Formulation
Maximize
T
 q ji  TQi 
j :iN j
e q
jN i
ij ij
s.t.
 qik  0
qik
kN i
 Ei  0
qji
i
qij  0
Ei
i
TQi
eijqik
Note: The problem
looks like max-flow
problem, but it is not
because the link cost
qik changes with
different receivers
Ni
28
Problems with Off-line Scheduling
• The dynamics of a network is not taken into account
– The nodes are static does not mean the links are static
either
– What happens after the first node dies
• The energy consumption profile is not correct
– The energy consumption is related to nodes instead of
links
– The overhearing effect should be considered
• No existing routing is flow-based (again due to the
required adaptability to the network dynamics). So
this approach provides only an upper bound on the
network lifetime
29
Data MULES: Let the God Collect Data
Access points
=
+
MULES
Sensor nodes
Random walk
The problem: only the God knows when the collected data
can be delivered to the access points and when the
MULEs are going to show up to pick up the next bunch
of data
30
Exploiting the Predictability
Access points
Predictable
Observer
Sensor nodes
Pre-defined route
The problem:
• The speed does matter
• Why don’t they consider multi-hop? If a multi-hop
routing is adopted, how should it behave?
31
Why Controlled Mobility
• Adapt topology to network requirements
– More adaptation than possible with protocol parameter
configuration in static nodes
• Increase capacity
– Enhanced bandwidth
– Energy saving
• Repair faults
– Connect sparse networks
• Other benefits
– Improved localization, time synchronization, coverage,
calibration, security
32
Slides borrowed from Kansal et al. [26]
System Infrastructure
Controlled Mobile Element Used to Route Data
Access Point
Mobile Router
Data Sources
33
Slides borrowed from Kansal et al. [26]
System Hardware
MOTE
Static Node Hardware
MOTE
STARGATE
PACKBOT
Mobile Router
Hardware
34
Slides borrowed from Kansal et al. [26]
Energy and Bandwidth Advantages
Hop distance
to base
5
4
4
4
3
3
4
C
3
2
2
3
B
1
1
2
5
2
Multihop Routing
D
A
Mobile Infrastructure
• Relay traffic reduced (energy saving)
• Number of wireless error-prone hops reduced (enhanced
bandwidth)
35
Slides borrowed from Kansal et al. [26]
Problems
• Sacrificing latency for bandwidth and energy
– Complicated control scheme is necessary to adaptively
change the moving speed
– What happens if nodes do not have enough memory to
cache the data
• Lack of analytical results, especially the results on the
optimality of the movement trajectory
• Adaptive mobility requires sophisticated routing
protocol
36
Power Consumption Profile of
Low Power Radio
– CC1000 radio and B_MAC of Mica2 Motes
Common Impressions are WRONG
• Tx power consumption is significantly higher than that of Rx
Mode
Current (mA)
Rx
7.00
Tx (dBm)
-20
…
0
…
6
…
3.70
…
8.50
…
13.80
…
• No power consumption if no receiving or transmission
happens
38
• A node consumes Rx power only when receiving its own packet
Tx Power vs. Rx Power
• Tx power determines the transmission coverage
30%
60%
90%
– There is no deterministic Tx range for a given Tx power
• Rx power is always fixed
• The radio can already achieve a remarkable Tx
coverage with Tx power < Rx power
39
Zoom in Rx Power – Idle Listening
• A power-on radio consumes energy constantly because
of idle listening, i.e., listen to an idle channel
• It turns out that all transmissions and receptions are free
• This result invalidates numerous research efforts
– For example
40
Solutions to Reduce Idle Listening
• Solution 1: Coordinated Sleeping (S-MAC of UCLA)
Listen
Sleep
Listen
Sleep
Listen
Sleep
Listen
Sleep
Synchronization between two neighbor nodes
– Distributed synchronization consumes energy!
• Solution 2: Preamble Sampling (B_MAC of
Berkeley)
Receiver
Sender
Sleep
Sleep
Preamble
Data Rx
Data Tx
41
Zoom in B_MAC – Duty Cycle
• Duty cycle (the percentage of radio power-on time)
is tunable!
Receiver
Sender
Sleep
Sleep
Preamble
Data Rx
Data Tx
CheckInterval
– Preamble length ≥ Sleeping time
– Long sleeping time trades transmission latency for low
power consumption (suitable for sparse transmission)
– A long preamble increases the power consumption of all
nodes in the sender’s transmission coverage due to
overhearing
42
Zoom in Rx Power – Overhearing
• A node has no knowledge about the destination of a
packet before it fully receives the packet!
Nonreceiver
Sender
Data Rx
Data Tx
Preamble
– An unlucky non-receiver spends energy to receive a long
preamble and the following packet
• Current solution: RTS/CTS
Nonreceiver
C
Receiver
Data Rx
43
Sender
Preamble
R
Data Tx
Problem with RTS/CTS
• Transmitting RTS/CTS consumes energy, and also
increases the complexity of MAC protocol
• When the duty cycle is low, the length of preamble is
significantly longer than the packet length, so using
RTS/CTS does not help too much
Maybe you will have a magnificent
solution 
44