Introduction to Wireless Networks
Michalis Faloutsos
1
What is an ad hoc network
A collection of nodes that can communicate with each
other without the use of existing infrastructure
Each node is a sender, a receiver, and a relay
There are no “special nodes” (in principal)
• No specialized routers, no DNS servers
Nodes can be static or mobile
Can be thought of us: peer-to-peer communication
2
Example: Ad hoc network
Nodes have power range
Communication happens
between nodes within
range
3
Some Introductory Things
The MAC layer 802.11
Typical Simulations
The routing protocols
TCP and ad hoc networks
4
What Is Different Here?
Broadcasts of nodes can “overlap” -> collision
How do we handle this?
A MAC layer protocol could be the answer
If one node broadcasts, neighbors keeps quite
Thus, nearby nodes compete for air time
This is called contention
5
Contention in ad hoc networks
A major difference with wireline networks
Air-time is the critical resource
Fact 1: connections that cross vertically interfere
Fact 2: connections that do not share nodes
interfere
Fact 3: a single connection with itself interferes!
6
Example of contention
E
F
G
H
B
A
C
D
Yellow connection
bothers pink
connection
Yellow bothers itself
When A-E is active
• E-F is silent
• F-G is silent (is it?)
7
The 802.11 MAC protocol
RTS
C
RTS
A
B
CTS
CTS
D
Introduced to reduce collisions
Sender sends Request To Send (RTS): ask permission
Case A: Receiver gives permission Clear To Send (CTS)
Sender sends Data
Receiver sends ACK, if received correctly
Case B: Receiver does not respond
• Sender waits, times out, exponential back-off, and tries again
8
Why is this necessary?
RTS
A
C
B
CTS
D
A: RTS, and B replies with a CTS
C hears RTS and avoids sending anything
• C could have been near B (not shown here)
D hears CTS so it does not send anything to B
9
Some numbers for 802.11
Typical radius of power-range: 250m
Interference range: 500m
• At 500m one can not hear, but they are bothered!
RTS packet 40 bytes
CTS and ACK 39 bytes
MAC header is 47 bytes
10
Typical Simulation Environment
A 2-dimensional rectangle
Fixed number of nodes
Static: uniformly distributed
Dynamic: way-point model
• Pick location, move with speed v, pause
Power range: fixed or variable
Sender-receivers uniformly distributed
11
Various Communication Paradigms
Broadcasting:
• one nodes reaches everybody
Multicasting:
• One node reaches some nodes
Anycasting:
• One node reaches a subset of some target nodes (one)
Application Layer protocols and overlays
• Applications like peer-to-peer
12
Layered and Cross Layer Protocols
application
application
transport
transport
Network
Network
Link
Link
physical
physical
Layering:
• Modular
• Isolates details of each layer
Cross Layer:
• Information of other layers
is used in decisions
• Pros: efficiency
• Cons: deployability and
compatibility
13
Example: application layer multicast
Source unicasts data to some
destinations
Destinations unicast data to
others
Pros: easy to deploy, no need
to change network layer
Cons: not as efficient
14
Example: application layer multicast II
s
A
Members need to make
multiple copies
• It would happened anyway
Link A B gets two packets
B
• Similarly in wireline multicast
Node B sends and receives
packet 4 times
15
Some major assumptions
The way-point model is a good model for
mobility
Homogeneity is a good assumption
Links are bidirectional: I hear U, U hear me
Uniform distribution of location is good
802.11 will be used at the MAC layer
Space is two dimensional
16
Some “proven” claims
The smallest the range, the better the throughput
Mobility increases the capacity of a network
A node should aim for 6-7 neighbors
We can challenge these claims
17
End of Introduction
Resources:
• Google:
• Citeseer: http://citeseer.nj.nec.com/cs
• C. Perkins book: Ad Hoc Networking
18
Modeling Contention
(based on Nandagopal et al MOBICOM 200)
Seminar 260
Michalis Faloutsos
19
Problem: Find Hotspot in a graph
Given a graph
and sourcedestinations
Where is the
bottleneck?
Or how much
bandwidth can
each connection
have?
20
Solution: Find areas of contention
Intuition
Step 1: create graph “range connectivity”
Step 2: create graph of flows (route flows on
graph)
Step 3: find which flows contend for airtime
(find areas where only one flow can be active)
21
Clarification: interference
When C->D
A-B, B-C, D-E, E-F can
not be active!
22
Clarification: Dual graph
edge
Each edge becomes a
node in G’
An edge exists between
two nodes in G’ iff
the edges have a
common node
Interference
23
In more detail
1.
2.
3.
4.
Find topological graph
Find dual graph: edges ->
nodes
Consider “interference”
between non adjacent edges
Find Maximal cliques
24
Contention Modeling: conclusion
Elegant approaches and tools are available
The realism of the modeling must be considered
Do not over-generalize results when heavy
assumptions have been made
25
Considering Connections
If we know which pairs want to communicate,
we consider only these flows as contenders
Routing could be independent of contention of
an area
If routing is contention aware, then we have a
closed loop system:
• Routing -> contention -> routing -> ….
26
Question: what is optimal routing?
Given a graph, source-destination pairs
How do I route the flows to minimize
contention?
What happens if I do not know the connections
ahead of time (online version of problem)?
27
Modeling the Physical Channel
There are several ways depending on degree of
accuracy
Binary, simplified:
• in one prange you communication
• In two prange you interfere but do not communicate
28
Considering the power: path loss
P_R: received power
P_t: transmission power
d: distance
alpha: constant
29
The physical model
Pi
Noise + SUM_k Pk
=
Node Y hears node i, iff received power of i is
above a threshold beta
• Needs to rise above noise and other transmissions
30
A more optimistic channel model
Node Y hears i, if i is the “loudest”
Interference from other nodes: per pair comparison
Delta>0 is a protocol specified “guard zone”
31
Channel Modeling: Conclusion
Several different models
You need to find and justify the model you use
32
Topology Control
We cannot always control the mobility
We can control the network topology
• Power control
• Deciding to ignore particular neighbors
From a given graph G of possible connections
we keep a subset G’ of these connections
What is good topology?
…
33
Topology Control Metrics
What is good topology?
• Energy efficiency,
• Robustness to mobility,
• Throughput - capacity
34
35
Topics Of Interest - Wireless
Characterizing the ad hoc topology
• A snapshot
• Its evolution
Mobility
• Realistic mobility models
Effect of topology/mobility in
• Routing
• Multicasting in ad hoc networks
36
Topics of Interest - Wireline
Generating a realistic directed graph
Reducing a real (directed) graph to a small
realistic
Survey on graph generation models
Measuring the Internet topology
• Router level
• AS level
37
We need to model contention
First the obvious
• Adjacent edges
Second, one edge away, considering RTS CTS
Third, interference (500m instead of 250m)
• Modeling issue
38
Typical “Errors”
Mobility:
• too slow or too fast
• Mobility speed may not be the expected
Homogeneity may “hide” issues
• Few nodes are responsible for most traffic
• Some spots are more popular than others
Power range is too large for the area
• Ie radius 250m, a grid of 1Km -> one broadcast covers
“half” the area
39
What’s the problem?
There is no systematic way to model and
simulate such networks
No clue what are the right assumptions
Not sure how the assumptions affect the results
40
Consequences
Simulation results are
•
•
•
•
Meaningless
Unrepeatable
Incomparable between different analysis
Prone to manipulation
Claim: give me any statement, I can create
simulations to prove it
41
What Will We Do Here?
Identify assumptions
• Some of them are subtle
Characterize the scenarios
Study their effect on the performance results
42