Transcript PPT

CMPE 257: Wireless
Networking
SET 2:
Models, Limits,
Architectures, and Logic in
Wireless Ad Hoc Networks
Spring 2005
UCSC
CMPE257
1
Conferences and Journals







MONET, WINET, IEEE JSAC
IEEE Trans Mobile Computing
IEEE Trans Wireless Comm.
ACM Mobicom, ACM Mobihoc
ACM Sensys, ACM mobisys
IEEE SECON, IEEE MASS, IEEE ICNP
IEEE Infocom, IEEE WCNC, IEEE
Globecom
Spring 2005
UCSC
CMPE257
2
Emergence of Sensor and Ad
Hoc Networks
sensor and ad hoc networks
technology
Spring 2005
packet radio
UCSC
CMPE257
applications
3
Technology Push

CMOS: Far smaller chips


Micro sensors


Ability to sense everything, everywhere, all the time (e.g.,
motion, light, vibration, pressure, humidity, …, even IDs)
Wireless spectrum: availability of large amounts
of unlicensed spectrum


Small and inexpensive processing and storage
Always-on connectivity
Radios: UWB, software radios, beam-forming,
successive interference cancellation

Better ways to use available spectrum
Spring 2005
UCSC
CMPE257
4
Technology Push (2)

Batteries and low-power chip and architecture
designs


Hardware and software (e.g., peer-to-peer,
caching, multi-modal UI)


Long-term unattended operation
Ability to make good trade-offs for processing, comm., energy
use, application effectiveness, and interaction with the world
Robots

Ability to integrate robots and routers (“robo-router”) to augment
the usefulness and capacity of wireless networks
(e.g., instruct a robo-router to move to a location to collect data,
reconnect a network, or improve the network throughput)
Spring 2005
UCSC
CMPE257
5
Application Push:
Netcentricity
Computing (networking, processing and
storage) is everywhere and invisible
 Long-lived applications
 Applications in highly disruptive environments
(e.g., in the battlefield, monitoring of
hazardous phenomena)
 Applications without end-to-end connectivity
 Real-time computing while interfacing with the
real world
 Human-to-content interaction
Spring 2005
UCSC
CMPE257
6
Example MANET:
i
g
C1
R
k
H1
f
p
h
j
e
m
c
b
d
R
C2
H2
a
What happens in a typical MANET
protocol stack?
Spring 2005
UCSC
CMPE257
7
What’s Wrong with Ad Hoc Nets?

Network control is based on algorithms
designed for graphs with point-to-point
links in which interference is confined
to each link.
Spring 2005
UCSC
CMPE257
8
Radio Channel Propagation
Effects




Spring 2005
UCSC
CMPE257

Diffraction
Reflection
Scattering
Time spread
Time-varying
9
Signal Strength at The
Receiving Antenna(s)
Path loss
Small-scale fading
S(dB)
shadowing
Receiver
Sensitivity
Distance
Spring 2005
UCSC
CMPE257
10
Interference Is Network-Wide!


No centralized control is
feasible
“Known” problems:




Topology is not a
“Boolean function”:



Spring 2005
UCSC
“Hidden terminal”
“Exposed terminal”
Scheduling
Quality of links depends
on activity of other nodes
RF propagation effects
Thermal/background
noise
CMPE257
11
What’s Wrong with Ad Hoc Nets?

Network architecture is based on the
Internet architecture.
Ad hoc network is viewed either as a subnet or a
“leaf” component of the Internet.
 Destinations in routing tables are hosts or
groups of hosts (nets).

Spring 2005
UCSC
CMPE257
12
The Beginnings of Protocol
Layering
A “remote backbone”
HOST
IMP
HOST
IMP
IMP
application



application
IMP
Routing within ARPANET is transparent to hosts
attaching to the ARPANET
The actual “customers of ARPANET” are the hosts!
People and processes are an after thought.
Spring 2005
UCSC
CMPE257
13
IP Internet today
R
R
R
R
R
R
R
R




R
A single path to each destination.
Topology is a given; link costs are static; end-to-end connectivity exists.
No control over what traffic is allowed on a given link (no usage policies)
Can’t select path with custom performance characteristics

All traffic must use resource rich path
Spring 2005
UCSC
CMPE257
14
IP Internet today
R
R
R
R
R
R
R
R



R
Still a “remote” backbone connecting hosts!
Host processing remains far removed from router processing (how
information is distributed and how resources are shared).
Usage policies must be implemented outside the routing system
Spring 2005
UCSC
CMPE257
15
Disruption-Tolerant Networks?
Remember: In the Internet model, topology is a given; link costs
are static; end-to-end connectivity exists.
Very short range
z
z
End-to-end connectivity need not
exist!
Spring 2005
UCSC
CMPE257
16
Disruption-Tolerant Networks?
z
Consider routes as functions of space and time;
exploit longer-term storage
Spring 2005
UCSC
CMPE257
17
What’s Wrong with Ad Hoc Nets?

No good handle on selfconfiguring networks.
Many proposals on “sufficient”
conditions to ensure loop freedom,
but without ensuring that the
protocol signaling always ensures
that such conditions are satisfied!
Spring 2005
UCSC
CMPE257
18
Current Routing in MANETs

Pro-active routing protocols


OLSR, DSDV, and STAR.
On-demand routing protocols

Source routed data packets



e.g., DSR
Not good in very large nets (source route is brittle)
Use routing invariants to perform hop-by-hop loop-free
routing (e.g., sequence numbers)


Typical case:
• AODV uses destination-based sequence numbers.
• No loop can exist because nodes can only trust higher
sequence number!
Has the IETF really covered all the bases?
Spring 2005
UCSC
CMPE257
19
Routing Using Destination
Sequence Numbers

Ad-hoc On-demand Distance Vector
Protocol (AODV).
Loop-freedom by ordering non-decreasing
destination sequence numbers towards a
destination.
 Performance suffers due to nodes requiring
sequence number ‘resets’ from the destination
on link failures.


Termination in the presence of state
loss, and node failures cannot be
guaranteed.
Spring 2005
UCSC
CMPE257
20
Fixed Spectrum Assignment
i
g
C1
R
k
f
p
h
j
e
m
c
b
d
R
C2
a
Poor connectivity
Spring 2005
UCSC
CMPE257
21
Fixed Spectrum Assignment
i
g
C1
R
k
f
p
h
j
e
m
c
b
d
R
C2
a
Too much interference:
Nodes are forced to use parts of spectrum
that are accessed by too many nodes.
Spring 2005
UCSC
CMPE257
22
Goal: Good Connectivity and Controlled
Interference
i
g
C1
R
k
f
p
h
j
e
m
c
b
d
R
C2
a
How should nodes elect which links to use with peers?
Should decisions be “cluster” or “node” based, local or
network-wide?
What signaling should be applied?
Spring 2005
UCSC
CMPE257
23
Good Connectivity and Controlled
Interference:
i
g
C1
R
k
f
p
h
j
e
m
c
b
d
R
C2
a
How should nodes monitor spectrum?
What is the impact of physical-layer parameters,
including node location?
MAC --> MASC (Medium Access and Selection Control)
Spring 2005
UCSC
CMPE257
24
What’s Wrong with Ad Hoc Nets?

No real clue on how users and protocol
stack should use available resources
efficiently.

Example, what if we have many links
between the same two nodes?
Spring 2005
UCSC
CMPE257
25
Policy-Based Routing
wired or optical
i
g
C1
R
k
f
p
h
j
e
m
c
b
d
R
C2
a
Spectrum agility means far richer connectivity.
We have a very different type of ad hoc nets: Any pair
of nodes can be connected by multiple links.
Spring 2005
UCSC
CMPE257
26
Policy-Based Routing
wired or optical
i
g
C1
R
k
f
p
h
j
e
m
c
b
d
R
C2
a
At certain locations, nodes may not be allowed to use
portions of the spectrum, or portions of the spectrum
may suffer too much interference.
Spring 2005
UCSC
CMPE257
27
Policy-Based Routing
wired or optical
i
g
C1
R
k
f
p
h
j
e
m
c
b
d
R
C2
a
To be effective, routing in spectrum agile networks
must be done with QoS, admin., and BW-use
constraints
Using location information is very important!
Spring 2005
UCSC
CMPE257
28
Not All Nodes and Traffic Are
Created Equal!
Most communication is
multipoint and for particular
purposes
command center
Image from
sensor
Spring 2005
UCSC
CMPE257
29
P-B Routing and Scalability of
Networks
D
Conventional
close straight
line path
S
Path of least
interference and
least resistance
subject to constraints
How can we reduce interference subject to multiple constraints
(spectrum available, power consumption, e-t-e delays,
bandwidth requirements…)?
Exploit diversity (user, space, time, code, freq), location
information and cross-layer optimization
Spring 2005
UCSC
CMPE257
30
What’s Wrong with Ad Hoc Nets?


Usage policies?
Security?
Who do I have to trust to establish an ad hoc
guest wireless group in a host infrastructure?
 Can I exploit available resources to enhance
security?


Who plays system administrator for the
embedded Internet?
Spring 2005
UCSC
CMPE257
31
Wireless Networks Are Very
Different than Wireline Netwoks
MAC and etiquettes establish links; need multicast
group affiliations and routes to destinations of flows for
better scheduling of spectrum
topology control
determines nodes &
links that can be
used for certain
functions; needs
links for interferencefree transmission of
control packets, and
dissemination of
neighborhood data
Spring 2005
S
Scalable &
Efficient
Network Control
T
R
routing needs links
for transmission of
control packets;
packet forwarding
needs links for
transmission of data
packets
Signaling to support
functions should not
be redundant
UCSC
CMPE257
32
Spectrum Agile Networking
CONSTRAINTS:
constraints


Spectrum: Spectrum
allocation rules
Traffic: Traffic engineering and
quality of service


Nodes: characteristics and
state of nodes in the network

Spring 2005
UCSC
Security and privacy would fall
here
e.g., power and storage
constraints
CMPE257
33
Role of Limits?
Spring 2005
UCSC
CMPE257
34
Recent Limits for Ad Hoc
Networks


Definition: A source-destination throughput of Λ(n)
bits/sec is feasible if every source node can send
information at a rate of Λ(n) bits/sec to its destination
for n total nodes in the network.
Gupta and Kumar [2000] (for static networks)

 n 

1
n  n 


 
 ( n)  

 0 and D(n)  
 

 n log( n) 
 log (n) 



Grossglauser and Tse [2001] (Multiuser diversity:
One-copy two phase packet relay to nearest neighbor
strategy for mobile networks)
(n)  1 and D(n)  (n)
Spring 2005
UCSC
CMPE257
35
A Different Model (SECON 04 paper)
Uniform Mobility
Model
Single-copy forward
 steady state distribution
of nodes is uniform.
Multi-copy forward
r0
r0
n total
users
r0
t'  t
Only one relay (the nearest)
looking for destination
Spring 2005
UCSC
n total
users
r0
First relay reaching destination (and
not necessarily the nearest) delivers the packet
(More than one relay looking for destination)
CMPE257
36
Enforcing One-copy Delivery
r0
p
n total
users
r0
i
j
Handshake
(check SN)
d(i)
k
Phase 2
Phase 1
Time-to-Live threshold (TTL) forces packets in
nodes p and k to be dropped.
Spring 2005
UCSC
CMPE257
37
Results in SECON 04 Paper

A multi-copy one-time relay strategy that
attains the Θ(1) throughput but provides
bounded delay for finite number of nodes n.
 Computes the interference effect and showed that
 ,   2
SIR n
 cte
 Presents an approximated formula for throughput as
a function of network parameters 1 1/  K 1 1  1  k 1/ 


e


1 
  e
k  0 k!   
 Computes a delay relationship between single-copy
K
and multi-copy relay strategy d  1 log 

K
Spring 2005
UCSC
CMPE257


 d 
K

1

e


38


Role of Limits
We need to understand fundamental
performance limits for any protocol stack of
the ad hoc networks we need.

Performance limits are meaningful after we have
established what an ad hoc network should be.
 Hence, limits are meaningful only within an
architectural context.
 This brings us back to the prior slides!


Examples: End-to-end connectivity is assumed for
information exchange, and source-destination pairs
compete with one another
Spring 2005
UCSC
CMPE257
39
Role of Limits (Cont.)

Do we need end-to-end connectivity all the
time and can (should) we even try to
enforce it?



What is the actual lifetime of an average link in the
battlefield?
What is the impact of locality of reference
and mirroring of content near its demand
points?
Why should we assume that sourcedestination pairs compete with one another?

What about SIC and other techniques?
Spring 2005
UCSC
CMPE257
40
Role of Limits (Conc.)

Why shouldn’t we view storage as part
of the communication bandwidth
available? (turn store and forward into
store-carry-forward)



Early examples: Single and multi-copy relay
schemes following Grossglauser and Tse’s work
If routes are plans in time and space,
what does it mean to be “connected”?
What is the capacity of a
“disconnected” network?
Spring 2005
UCSC
CMPE257
41
Modeling The
Impact of
Physical Layer?
Spring 2005
UCSC
CMPE257
42
Role of Interactions in The
Modeling of Ad Hoc Networks





Need to model specific protocol stacks to predict
performance.
Must include physical layer aspects directly into
the behavior of MAC protocols (and protocol
stack)
Must consider interdependencies among nodes
given by radio-based topology
Per-node performance
Model must be scalable (faster than simulation)
Spring 2005
UCSC
CMPE257
43
Modeling: Previous Work



Single-hop or “weak interactions” approach
Scheduling rates modeled as independent
Poisson processes
Packet lengths:


Exponentially distributed
Independent at each transmission attempt:





back-off schemes ignored!
Instantaneous acknowledgments
Error-free links
Assumptions on spatial distributions (e.g.,
Poisson)
Result: Heavy reliance on simulations
Spring 2005
UCSC
CMPE257
44
Modeling Goal:
Reflect Interactions between PHY
and MAC Layers

Focus on the essentials of PHY and MAC layers:




PHY: to ensure that frames are received correctly
MAC: scheduling discipline to access the channel
PHY/MAC dynamics tightly connected
PHY/MAC interactions depend on connectivity
among the nodes:


Network topology is key!
Model each layer’s functionality probabilistically:


PHY: probability of successful frame reception
MAC: transmission probability (scheduling rate)
Spring 2005
UCSC
CMPE257
45
“Generic” Modeling Approach

PHY:


MAC:


The probability of successful reception of a data packet
and its acknowledgment, based on effect from all
transmissions (which depend on scheduling by the
MAC) and PHY parameters
Scheduling rates based on feedback from the PHY
regarding the success of transmissions
Topology:


Consider the effect of all nodes based on where they
are and their transmissions
Simplify the problem taking advantage that MAC
protocol will tend not to schedule transmissions when
feedback from the PHY indicates unsuccessful
transmissions
Spring 2005
UCSC
CMPE257
46
Impact of Physical Layer


Consider the effect of network-wide
interference
Signal-to-interference-plus-noise density ratio:
SINRir 
Pi r Li
r
2

P


 j j r
jVr , j  i
where:
Pkr  signal power perceived at r for signal sent by k Vr
Li  spreading gain
 r2  background/thermal noise power at r
1,
j  
0,
Spring 2005
if j transmits at the same time
otherwise
UCSC
CMPE257
47
Impact of Physical Layer

Successful frame reception probability

Let Cir denote a set of potential interferers:
qir  P successful frame reception


  P successful frame reception, Cir  cikr
k

  P successful frame reception Cir  cikr
k
  
  f cikr P Cir  cikr
k
Assume:


 PC
r
i
 cikr



  1   
P Cir  cikr 
m
mcikr
n
ncikr
Note that scheduling rates (taus) are given by the MAC
layer!
Spring 2005
UCSC
CMPE257
48
Impact of MAC Layer


Consider a reliable delivery service
MAC as a stochastic dynamic system:

Feedback: successful transmission probabilities
   
 
qi  qir qri   f cikr f crli P Cir  cikr P Cri  crli
k




l
Output: scheduling rates
 i , i V
Steady-state operation (under saturation):
 i  hi qi , i V
In reality, MAC’s operation is a time-varying system, and
scheduling rates are also functions of packets in buffers.
Spring 2005
UCSC
CMPE257
49

Impact of Topology:
Linearization
First-order approximation of h q , withhi 0  0
i
i
(because MAC will tend not to schedule
transmissions when transmissions are not
successful):
 i  hi qi   aqi , where

Keep term with the highest SINR:
qi  f cir0 f cri 0  1   j   1   k 
   
jVr

a  hi' 0
kVi
If a  1,


qi   i 1   aqk , where  i  f cir0 f cri 0
 k  Vr  Vi 
  
Spring 2005
UCSC
CMPE257
50
Linear System

Linear system:
 0 12

 21 0
Φ  31 32


n1 n 2

I  Φq  π
13  1n 
23  2 n 
a i , if j  Vi  Vr
0  3n , with ij  


n 3 

 
0 
 0,
otherwise
Transmission prob. vector: τ  aI  Φ π
Spring 2005
1
UCSC
CMPE257
51
Application:
Modeling IEEE 802.11 DCF in
Multihop Ad Hoc Networks

M. Carvalho and J. J. Garcia-Luna-Aceves, “Delay Analysis
of IEEE 802.11 in Single-Hop Networks,” Proc. ICNP,
Atlanta, 2003.



Node’s service time as a function of channel state probabilities
Model extension: finite back-off operation
G. Bianchi, “Performance Analysis of the IEEE 802.11
Distributed Coordination Function,” IEEE JSAC, 2000.

Functional form 
Spring 2005
 hq  (single-hop, ideal channel conditions)
UCSC
CMPE257
52
Model Validation:
Simulation Setup




Qualnet simulator (v3.5)
Radio Channel Model: “Two-ray”
Standard IEEE 802.11 DCF parameters
IEEE 802.11 (PHY):










Direct sequence spread spectrum (DSSS)
DBPSK at 1Mbps
Radio range: 200 m
Carrier sensing range: 400 m
Packet reception model: BER
Independent bit errors per frame
Area: 1000 x 1000 m
Nodes randomly placed in the terrain (but connected)
Fixed packet sizes: 1500 bytes
5 min. data traffic
50 trials corresponding to different initial transmission times

Spring 2005
UCSC
CMPE257
53
Model Validation:
Per-node Throughput
Scenario with 100 nodes
Spring 2005
UCSC
CMPE257
54
Prediction Error
Histogram over 10 random topologies
(100 nodes)
Sample topologies
Not wonderful,
but a good start!
Spring 2005
UCSC
CMPE257
55
Complementary of
Simulation Effort

Analytical model is much faster:

Simulation setup:








Platform: Sun blade 100 SunOS 5.8
50 seeds
100 nodes
5-min data traffic
Total time: 16.41 hours
Analytical model: 0.44 seconds in Matlab 6.0
Analytical model is 105 faster than simulations!
Scalable simulations can address
entire protocol stack and complex
scenarios!
Spring 2005
UCSC
CMPE257
56
Role of logic?
Spring 2005
UCSC
CMPE257
57
Logic in Ad Hoc Networks




The logic is in the signaling used to control the networks.
Goal: Self-organizing and scalable ad hoc networks
Protocol layers operate in isolation (e.g., routing, MAC
scheduling and topology control are mutually
independent)
Too many assumptions are being made



Single channel, all nodes are equal, etc.
Many proposals on “sufficient” conditions to ensure loop
freedom exist, but without ensuring that the protocol
signaling always ensures that such conditions are always
satisfied!
Too much reliance on global variables (e.g., hold down
timers after reboot)
Spring 2005
UCSC
CMPE257
58
Example: Routing in
MANETs

Pro-active routing protocols



Examples, OLSR, DSDV, and STAR.
Too much signaling since not all nodes need to talk to all other
nodes with same likelihood.
On-demand routing protocols

Source routed data packets



e.g., DSR
Not good in very large nets (source route is brittle)
Use routing invariants to perform hop-by-hop loop-free
routing (e.g., sequence numbers)

Typical case:
• AODV uses destination-based sequence numbers.
• Hope: No loop can exist because nodes can only trust higher
sequence number!
Spring 2005
UCSC
CMPE257
59
Routing Using Destination
Sequence Numbers

Example: Ad-hoc On-demand Distance Vector Protocol
(AODV).



Termination in the presence of state loss, and node
failures cannot be guaranteed without a global
parameter.


Loop-freedom by ordering non-decreasing destination sequence
numbers towards a destination.
Performance suffers due to nodes requiring sequence number ‘resets’
from the destination on link failures.
Recent example: “Wait until none of the network nodes can
possibly use node is question in their paths to a destination.”
How long is it safe to wait after state loss to ensure
no counting-to-infinity?
Spring 2005
UCSC
CMPE257
60
Logic in Ad Hoc Networks

Need to consider network architecture
and fundamental limits!
Spectrum agility for scaling
 Role of store-carry-forward and routes as plans
in space and time
 Modular signaling that works well for tiny and
very large networks, none or multiple policies,
and without anyone having to choose
parameter values.
 Cross-layer interaction

Spring 2005
UCSC
CMPE257
61
In summary…
ARCHITECTURES
LOGIC:
Self-Organizing, scalable
LIMITS
ANALYTICAL MODELS
& SIM
Spring 2005
UCSC
CMPE257
62