Roofnet
Free Wireless Internet in Cambridge
ESD 342
Presented by: Derek Rayside, Yingxia Yang
Jennifer E Underwood
Supervised by: Joel Moses
Roofnet Revisited
The goal of Roofnet project is to provide
broadband wireless Internet access to
users in Cambridge.
Data Inconsistencies and Issues
Coordinate data in the SIGCOMM2004 paper
supplementary information:


Incomplete (some nodes in traffic data not in
coordinate file)
Inconsistent (some lat-long coordinates do not match
up with map in paper)
Time stamp data not synchronized


Some packets arrived before they were sent
Couldn’t calculate transmission time
Estimated based on s/n ratio and nominal bit rate
Gateway nodes unclear…
Traffic data arranged by “experiment”


Attempted bit rates: 1, 2, 5.5, and 11 Mbps
One node sends while others listen and record
Data Inconsistencies:
Coordinates and Node ID
Revised Project Goals
Analyse network topological properties
and variation in connectivity strengths as
attempted data rate increases

Have analyzed for aggregate data
Analyse performance of periphery nodes

Indicated as problem by Roofnet group
Analyse robustness

In progress
Analyse congestion (if time)
Periphery: where’s the edge?
Connectivity:

Few partners
Bandwidth

Low s/n
Geography

Physical distance
All 3 are different
Connectivity Periphery:
≤ 4 incoming/outgoing links
Nodes with Red X’s
are the nodes
identified as being “on
the margin”
No data to locate
these nodes
geographically
These send but do
not receive signals…
Basic Quantitative Analysis on
Aggregate Data
Nodes: 41
Edges: 562
Average degree: 562/41=13.7
Maximal out-degree = 27
Maximal in-degree = 26
Average path length: 0.3760
Harmonic path length: 5.5962
Clustering Co-efficiency: 0.5625
Centrality:

Degree centrality
Degree centrality for out-degree: 34.063%
Degree centrality for in-degree: 31.500%

Betweenness centrality
Network Centralization Index = 9.19%
Degree distribution
e
e
r
g
e
d
x
e
t
r
e
v
degree distribution
30
25 3511
22
3817
41
471215
16
26
138
24
27
3139
10
33
13
2814
34
18
23
2032
36
26
21
20
15
10
5
0
0
10
20
30
529
37
9
30
19
2540
40
50
vertex ordered index
e
e
r
g
e
d
8
m
a 6
r
g
o
t 4
s
i
h
2
0
y
c
n
e
u
q
n
e
o
ir
tf
u
b
i
r
t
s
i
d
0
5
10
15
20
25
30
10
0
10
0
10
1
10
2
degree
10
2
e
v
1
i 10
t
a
l
u
m
u
c 10 0
10
0
10
1
degree
10
2
Prestige and Acquaintance
Based on aggregate data
Authority: not only referred to by many nodes, but
also by many Hubs. (measurement: prestige)
Hub: not only refers to many nodes, but also to many
Authorities. (measurement: acquaintance)
Asymmetry
s/n not symmetrical
X can talk to Y
Can Y talk to X?
some dramatic
differences
76 one-way edges
Most links are low quality
Connectivity
& S/N Quality
<4db
< 20db
> 20db
connectivity ≠
bandwidth
“Quality” Links: Geographical
Nodes with Red X’s
had inconsistent
data: map represents
best-guess
location/node ID
Strongly-Connected
Components
Subgraph where every node can reach
every other node
Collapse into “meta-node”
All links
“Quality” links (>4db s/n)
Future steps
Finish analyzing variation across
experiments
Finish robustness analysis

Random and targeted failed nodes
Finish analyzing periphery nodes
If time


Analyze congestion in OPNET
Evaluate political situation in Cambridge with
regards to deploying Roofnet
Back-up Slides
3-node Motifs
Full list includes 2 motifs
MOTIF NREAL NRAND
ID
STATS
108
NREAL
ZSCORE
NREAL
PVAL
UNIQ
VAL
CREAL
[MILI]
185
153.7+-10.1
3.08
0.000
5
56.66
742
601.0+-13.2
10.68
0.000
10
227.26
001
101
100
238
011
101
110
Had ALL 13 of the possible sub-graphs
Oops!
Three nodes only sending signals without
receiving any signals


23745, 43224, 26222
Interestingly, these nodes did not appear in
the coordinate data corresponding to the
Roofnet SIGCOMM2004 paper…
Nodes “on the margin” don’t have good
connectivity to the network


Appear to be periphery nodes on this basis
Define as those nodes with ≤4
incoming/outgoing links