FRACTEL: Building (Rural) Mesh
Networks with Predictable Performance
On the Feasibility of the Link
Abstraction in (Rural) Mesh Networks
Dattatraya Gokhale (Indian Navy),
Sayandeep Sen (U. Wisconsin-Madison),
Kameswari Chebrolu (IIT Bombay),
Bhaskaran Raman (IIT Bombay)
Work done at IIT Kanpur
Presentation at Infocom 2008, Phoenix, Apr 2008
LocalGateway:
gateway to
LDN
FRACTEL Deployment
wiFi-based Rural
data ACcess &
TELephony
Alampuram
Tetali
Point-to-Multi-Point
802.11 link-sets using
Sector Antennas
Kasipadu
LACN: LocalPoint-to-Point
ACcess Network
802.11 Links with
at one of the
Directional
villages (desired,
Antennas
not deployed)
Pippara
Ardhavaram
Kesavaram
Jalli Kakinada
19 Km
Korukollu
Polamuru
IBhimavaram
Tadinada
19.5 Km
Cherukumilli
Juvvalapalem
Need to support
real-time apps.
Bhimavaram
Landline: wired
gateway to the
Internet
Jinnuru
Lankala Koderu
LDN: LongDistance Network
(deployed, in the
Ashwini project)
Link Abstraction: Background
• Link abstraction:
Packet Error Rate (%)
– Either link exists
or does not
– That is, 0%
packet reception,
or ~100%
– Abstraction holds
in wired
networks
Link doesn’t exist.
Steep change in Error Rate
Link Exists
Negligible error rates
Average RSSI (dBm)
• Understanding link behaviour has implications on
– Network Planning, Protocol Design, Application Design
DGP, Roofnet, FRACTEL LACN
Typical
link
distances
Network
architecture
Environ
ment
Up to few
tens of
kms
High gain
directional &
sector
antennas on
tall towers or
masts
Rooftop
mesh
networks
(e.g.
Roofnet)
FRACTEL
LACNs
Longdistance
mesh
networks
(e.g. DGP)
Multipath
effects
SNR or
RSSI
External
interference
Link
abstra
ction
Rural
setting
studied in
depth
Effect not
apparent
Has strong
correlation
with link
quality
Affects links
performance
Valid
Mostly <
500 m
Mostly omnidirectional
antennas on
rooftops
Dense
urban
setting
studied
in-depth
Reported
as a
significant
component
Not useful
in
predicting
link quality
Reported as
not
significant
Not
valid
Mostly <
500 m
Would like to
avoid tall
towers
Rural,
campus,
residential
To be
determined
To be
determined
To be
determined
To be
determ
ined
DGP, Roofnet, FRACTEL LACN
Typical
link
distances
Network
architecture
Environ
ment
Up to few
tens of
kms
High gain
directional &
sector
antennas on
tall towers or
masts
Rooftop
mesh
networks
(e.g.
Roofnet)
FRACTEL
LACNs
Longdistance
mesh
networks
(e.g. DGP)
Multipath
effects
SNR or
RSSI
External
interference
Link
abstra
ction
Rural
setting
studied in
depth
Effect not
apparent
Has strong
correlation
with link
quality
Affects links
performance
Valid
Mostly <
500 m
Mostly omnidirectional
antennas on
rooftops
Dense
urban
setting
studied
in-depth
Reported
as a
significant
component
Not useful
in
predicting
link quality
Reported as
not
significant
Not
valid
Mostly <
500 m
Would like to
avoid tall
towers
Rural,
campus,
residential
To be
determined
To be
determined
To be
determined
To be
determ
ined
DGP, Roofnet, FRACTEL LACN
Typical
link
distances
Network
architecture
Environ
ment
Up to few
tens of
kms
High gain
directional &
sector
antennas on
tall towers or
masts
Rooftop
mesh
networks
(e.g.
Roofnet)
FRACTEL
LACNs
Longdistance
mesh
networks
(e.g. DGP)
Multipath
effects
SNR or
RSSI
External
interference
Link
abstra
ction
Rural
setting
studied in
depth
Effect not
apparent
Has strong
correlation
with link
quality
Affects links
performance
Valid
Mostly <
500 m
Mostly omnidirectional
antennas on
rooftops
Dense
urban
setting
studied
in-depth
Reported
as a
significant
component
Not useful
in
predicting
link quality
Reported as
not
significant
Not
valid
Mostly <
500 m
Would like to
avoid tall
towers
Rural,
campus,
residential
To be
determined
To be
determined
To be
determined
To be
determ
ined
DGP, Roofnet, FRACTEL LACN
Typical
link
distances
Network
architecture
Environ
ment
Up to few
tens of
kms
High gain
directional &
sector
antennas on
tall towers or
masts
Rooftop
mesh
networks
(e.g.
Roofnet)
FRACTEL
LACNs
Longdistance
mesh
networks
(e.g. DGP)
Multipath
effects
SNR or
RSSI
External
interference
Link
abstra
ction
Rural
setting
studied in
depth
Effect not
apparent
Has strong
correlation
with link
quality
Affects links
performance
Valid
Mostly <
500 m
Mostly omnidirectional
antennas on
rooftops
Dense
urban
setting
studied
in-depth
Reported
as a
significant
component
Not useful
in
predicting
link quality
Reported as
not
significant
Not
valid
Mostly <
500 m
Would like to
avoid tall
towers
Rural,
campus,
residential
To be
determined
To be
determined
To be
determined
To be
determ
ined
DGP, Roofnet, FRACTEL LACN
Typical
link
distances
Network
architecture
Environ
ment
Up to few
tens of
kms
High gain
directional &
sector
antennas on
tall towers or
masts
Rooftop
mesh
networks
(e.g.
Roofnet)
FRACTEL
LACNs
Longdistance
mesh
networks
(e.g. DGP)
Multipath
effects
SNR or
RSSI
External
interference
Link
abstra
ction
Rural
setting
studied in
depth
Effect not
apparent
Has strong
correlation
with link
quality
Affects links
performance
Valid
Mostly <
500 m
Mostly omnidirectional
antennas on
rooftops
Dense
urban
setting
studied
in-depth
Reported
as a
significant
component
Not useful
in
predicting
link quality
Reported as
not
significant
Not
valid
Mostly <
500 m
Would like to
avoid tall
towers
Rural,
campus,
residential
To be
determined
To be
determined
To be
determined
To be
determ
ined
DGP, Roofnet, FRACTEL LACN
Typical
link
distances
Network
architecture
Environ
ment
Up to few
tens of
kms
High gain
directional &
sector
antennas on
tall towers or
masts
Rooftop
mesh
networks
(e.g.
Roofnet)
FRACTEL
LACNs
Longdistance
mesh
networks
(e.g. DGP)
Multipath
effects
SNR or
RSSI
External
interference
Link
abstra
ction
Rural
setting
studied in
depth
Effect not
apparent
Has strong
correlation
with link
quality
Affects links
performance
Valid
Mostly <
500 m
Mostly omnidirectional
antennas on
rooftops
Dense
urban
setting
studied
in-depth
Reported
as a
significant
component
Not useful
in
predicting
link quality
Reported as
not
significant
Not
valid
Mostly <
500 m
Would like to
avoid tall
towers
Rural,
campus,
residential
To be
determined
To be
determined
To be
determined
To be
determ
ined
DGP, Roofnet, FRACTEL LACN
Typical
link
distances
Network
architecture
Environ
ment
Up to few
tens of
kms
High gain
directional &
sector
antennas on
tall towers or
masts
Rooftop
mesh
networks
(e.g.
Roofnet)
FRACTEL
LACNs
Longdistance
mesh
networks
(e.g. DGP)
Multipath
effects
SNR or
RSSI
External
interference
Link
abstra
ction
Rural
setting
studied in
depth
Effect not
apparent
Has strong
correlation
with link
quality
Affects links
performance
Valid
Mostly <
500 m
Mostly omnidirectional
antennas on
rooftops
Dense
urban
setting
studied
in-depth
Reported
as a
significant
component
Not useful
in
predicting
link quality
Reported as
not
significant
Not
valid
Mostly <
500 m
Would like to
avoid tall
towers
Rural,
campus,
residential
To be
determined
To be
determined
To be
determined
To be
determ
ined
Experimental Methodology
• Two kinds of environment
– Five locations on campus, One village location
• One transmitter, Multiple receiver positions
• Broadcast 6K 1400 byte pkts with 20ms gap.
• Hardware same as in DGP study, Roofnet study
Results: Err Rate vs RSSI
• For Interference prone positions
– Intermediate error rates
• For Interference free positions
– If RSSI > Threshold
Error rates are low and stable
Data Rate: 1Mbps
MIT Roofnet: Conclusions
• No correlation between SNR and Error Rate
– Loss rate high at high SNR
• No correlation between lost packets and foreign
packets observed
• Introducing delay spread causes high loss rates
Multipath induced delay spread and not
interference is the culprit
MIT Roofnet data: Fresh Analysis
Data Rate: 1Mbps, Average RSSI > -80 dBm, 80% >Error Rate > 20%
MIT Roofnet data: Fresh Analysis
Noise band as high as 16 dB, Ours ~ 2 dB
Data Rate: 1Mbps, Average RSSI > -80 dBm, 80% >Error Rate > 20%
MIT Roofnet data: Fresh Analysis
Roofnet Max Noise = -75 dBm, Ours = -94 dBm
Data Rate: 1Mbps, Average RSSI > -80 dBm, 80% >Error Rate > 20%
MIT Roofnet data: Fresh Analysis
Data Rate: 1Mbps, Average RSSI > -80 dBm, 80% >Error Rate > 20%
What is the cause of increased noise level?
Multipath does not cause high noise level
Is it Interference ?
Understanding Interference
1. Does interference on affect noise levels ?
2. Can pkts. loss be related to no. of foreign
pkts seen?
3. Can reported noise level be used to gauge the
level of interference?
4. Can we estimate the link performance based
on the average measured noise floor?
Understanding Interference
1. Does interference on affect noise levels ?
2. Does pkts. loss correlate with foreign pkts. ?
3. Use reported noise level to gauge level of
interf.?
4. Estimate the link perf. based on the avg.
measured noise floor?
Controlled Interference Expt
• Experimental Setup
• A and B Hidden Nodes to one another
• B’s power fixed at -75 dBm
• A’s power varied: -90, -85, -80, -75 dBm
Does Interference affect noise levels?
Controlled Experiment
Roofnet
Avg Received RSSI from A is
-85 dBm and B is -75 dBm
Noise extends right up to
-65 dBm
P1:
Interference causes noise level to be high and
variable
Can packet loss be related to foreign
packets seen?
Col-1
Col-2
Col -3
Col-4
Col-5
Col-6
Col-7
Col-8
Col-9
Expt
No
Src
Mean
RSSI
Loss
%
Mean
Noise
5
%-ile
95
%-ile
Noise
Band
Max
Noise
(dBm)
(dBm)
(dBm)
(dB)
(dBm)
-94
-94
-85
-85
9
9
-85
-84
(dBm)
2
2
A
B
-85.23 99.2 -92.53
-74.68 18.3 -89.34
As far as B’s packets are concerned:
Hidden node, receiver
B’s loss = 18.3%;
outside of interferer’s
A’s loss = 99.2%  4 foreign pkts /sec reception range
P2:
Packet loss high even though number of
observed foreign packets low
IN RANGE
NODE ‘A’
NODE ‘B’
NODE ‘R’
Node
A
• Senses
Medium,
Clear
• Starts TX
Node
B
Non-hidden-node case
• Senses
Medium
• Hears A
• Backs off
Node
R
No of
packets
seen low
P3: Packet loss can be low even though
number of observed foreign packets is high
P2
P3
Support Roofnet’s
observation but disprove
conclusion
Can the reported noise level be used
to gauge the level of interference?
Col-1
Col-2
Col -3
Col-4
Col-5
Col-6
Col-7
Col-8
Col-9
Expt
No
Src
Mean
RSSI
Loss
%
Mean
Noise
5
%-ile
95
%-ile
Noise
Band
Max
Noise
(dBm)
(dBm)
(dBm)
(dB)
(dBm)
-94
-94
-85
-85
9
9
-85
-84
(dBm)
2
2
A
B
-85.23 99.2 -92.53
-74.68 18.3 -89.34
Can the reported noise level be used
to gauge the level of interference?
Col-1
Col-2
Col -3
Col-4
Col-5
Col-6
Col-7
Col-8
Col-9
Expt
No
Src
Mean
RSSI
Loss
%
Mean
Noise
5
%-ile
95
%-ile
Noise
Band
Max
Noise
(dBm)
(dBm)
(dBm)
(dB)
(dBm)
-94
-94
-85
-85
9
9
-85
-84
(dBm)
2
2
A
B
-85.23 99.2 -92.53
-74.68 18.3 -89.34
Large Noise Band
– Instantaneous noise levels show variability
Can the reported noise level be used
to gauge the level of interference?
Col-1
Col-2
Col -3
Col-4
Col-5
Col-6
Col-7
Col-8
Col-9
Expt
No
Src
Mean
RSSI
Loss
%
Mean
Noise
5
%-ile
95
%-ile
Noise
Band
Max
Noise
(dBm)
(dBm)
(dBm)
(dB)
(dBm)
-94
-94
-85
-85
9
9
-85
-84
(dBm)
2
2
A
B
-85.23 99.2 -92.53
-74.68 18.3 -89.34
Actual Interference
Perceived Interference
– Noise levels reported differ from known level
P4:
On this H/W, gauging level of interference is error
prone
Can link performance be estimated
based on average noise floor?
From Roofnet
Data
P5: It is not possible to estimate the link
quality based on reported noise floor
Operating near the RSSI threshold
Variable Loss Rates
Village Location
Operating near the RSSI threshold
Variable Loss Rates
No Interference
Village Location
Operating near the RSSI threshold
Variable Loss Rates
No Interference
Village Location
What is the reason behind Intermediate error rates ?
RSSI Below Thresh.
Packet Error Rate
(%)
Operating near the RSSI threshold
Average RSSI (dBm)
Village Location
RSSI Above Thresh.
Packet Error Rate
(%)
Operating near the RSSI threshold
Average RSSI (dBm)
Village Location
Operating near the RSSI threshold
Small variation in RSSI  large
variation in error rates
Village Location
Cannot distinguish between links with loss rates between 0-100%
Cannot use routing metrics based on ETX or WCETT (going to be unstable)
Design Implications
• Link abstraction:
– Absence of external interference  can plan links
• Routing:
– Metrics to distinguish between links with
intermediate loss rates (e.g. ETX and WCETT)
are likely unstable
– Gauging interference can be error prone in
interference-aware routing
• MAC:
– CSMA/CA will lead to unpredictable loss rates,
due to self-interference and hidden node cases
Conclusion
• Multipath and delay spread:
– Not a likely factor in rural areas
– May or may not be the main culprit in urban areas
– Roofnet: extrapolates 900 MHz multipath measmts.
• Lesson: double-check measurements, analysis
• Open questions:
– 802.11g and 802.11a?
– Future of unplanned deployments?
• For further information on FRACTEL:
– http://www.cse.iitb.ac.in/~br/
Backup slides
Operating near the RSSI threshold
• In the absence of
external
interference
• Intermediate error
rates due to
operation in steep
region
• Small variation in
RSSI  large
variation in error
rates
Village Location
• Cannot distinguish between links with loss rates
between 0-100%
• Cannot use routing metrics based on ETX or WCETT
Interference free locations: RSSI
Stability
• Short-term stability (2-min expt): Mostly within 3-4 dB
Hardware
Quirk
Person standing
near antenna
Band: Difference between the 95%-ile and 5%-ile values of RSSI
Long-term Stability
• Band variation depends on environment but within
4dB in most cases
Location
Rx
Posn
LoS ?
Dur.
(hrs)
RSSI
95%-ile
(dBm)
RSSI
5%-ile
(dBm)
RSSI
Band
(dB)
(Y/N)
Apt
1
Y
48
-66
-69
3
Apt
2
N
(foliage)
48
-69
-77
8
Apt
3
N
(foliage)
48
-76
-82
6
Apt2Dorm
1
Y
24
-75
-77
2
Apt2Dorm
2
Y
24
-70
-71
1
Apt2Dorm
3
N
(foliage)
24
-79
-81
2
Design Implication: Link Abstraction
• Absence of external interference  can plan
links with predictable performance.
• Simplifies higher layer protocol design
considerably
RSSI threshold
-79dBm
Modified RSSI
Threshold Value
-75dBm
RSSI variation
3-4dB
Design Implication: MAC
• CSMA/CA unsuitable in multihop mesh
networks
• External interference can worsen CSMA/CA
performance
• Roofnet data indicates considerable sources at
interference range but not reception range
– RTS/CTS will not help in this setting
FRACTEL: Ongoing Work
• Scope for TDMA-based mesh network
– Lots of theory research, little in systems
• Systems issues in TDMA-based networks:
–
–
–
–
Interference mapping
Synchronization
Schedule dissemination, dynamic scheduling
Scaling
• For further information:
– http://www.cse.iitb.ac.in/~br/
Long Distance Networks (LDNs)
Link abstraction holds
Measurements on
the DGP network,
Kanpur, India
"Long-Distance 802.11b Links: Performance Measurements and Experience'',
Kameswari Chebrolu, Bhaskaran Raman, and Sayandeep Sen, MOBICOM 2006
Local Access Networks (LACNs)
• Prior Studies: MIT Roofnet study
– Outdoor WiFi mesh, Boston/Cambridge area
– Most links have intermediate loss rates, between
0% and 100%
– Multi-path (not external interference) is a major
cause of losses
– No Link Abstraction!
– Work around: design of appropriate routing
metrics
Experimental Methodology:
Environment
• Two kinds of environment
– Five locations on campus
– One village location
• Link lengths: 150-400m
• One transmitter position
• Up to 6 receiver positions
– Good – Avg. RSSI ≈ -70 dBm
– Medium – Avg. RSSI ≈ -75 dBm
– Bad – Avg. RSSI ≈ -80 dBm
FRACTEL Measmt. Study: IITK
Dense buildings, academic area,
student dormitories, campus housing,
several trees
Source: Google Maps
Understanding Interference
1. Does interference on affect noise levels ?
2. Can pkts. loss be related to no. of foreign
pkts seen?
3. Can reported noise level be used to gauge the
level of interference?
4. Can we estimate the link performance based
on the average measured noise floor?
FRACTEL Measmt. Study: Amaur
Dense buildings, 2-3 storey tall
Source: Google Maps
Experimental Methodology
• Hardware
– Laptops with Senao 2511CD Plus 802.11b
PCMCIA cards
Hardware same
– Antennas
as in DGP study,
Roofnet study
• Sector Antenna – 17 dBi
• Omni Directional Antenna – 8 dBi
• Software
– Linux – kernel 2.6.11
– Modified HostAP driver – ver 0.4.9
– Each experiment broadcasts 6000 1400 byte pkts
with 20ms gap at 4 different data rates
Controlled Interference Expt
• Experimental Setup
•
•
•
•
A: 1400-byte packets, 2ms interval, 11Mbps
B: 1300-byte packets, 2ms interval, 11Mbps
B’s power fixed at -75 dBm
A’s power varied: -90, -85, -80, -75 dBm
Can the reported noise level be used
to gauge the level of interference?
Col-1
Col-2
Col -3
Col-4
Col-5
Col-6
Col-7
Col-8
Col-9
Expt
No
Src
Mean
RSSI
Loss
%
Mean
Noise
5
%-ile
95
%-ile
Noise
Band
Max
Noise
(dBm)
(dBm)
(dBm)
(dB)
(dBm)
-94
-94
-85
-85
9
9
-85
-84
(dBm)
2
2
A
B
-85.23 99.2 -92.53
-74.68 18.3 -89.34
– Instantaneous noise levels show variability
– Noise levels reported differ from known level
P4:
On this H/W, gauging level of interference is error
prone