Transcript Slide 1

Architecture and Evaluation
of an Unplanned 802.11b
Mesh Network
John Bicket, Daniel Aguayo, Sanjit Biswas, and
Robert Morris
MIT Computer Science and Artificial Intelligence Lab
Presented by Anuradha Kadam
February 6, 2007
Outline





Introduction
Roofnet Design
Evaluation
Network Use
Conclusion
Introduction

Community wireless networks:

Multi-hop network with nodes in chosen locations
and directional antennas


Hot-spot access points to which clients directly
connect


Require well-coordinated groups with technical
expertise, result in good connectivity and throughput
Do not require much coordination to deploy and
operate, not as much coverage per wired connection.
Best characteristics of both network types.
Introduction





Unconstrained node placement
Omni-directional antennas
Multi-hop routing
Optimization of routing for throughput
Roofnet
Roofnet Design



37 nodes spread over
four square km
Each node hosted by a
volunteer
Most buildings are 3 or
4 story
Hardware




Node: PC, an 802.11b
card and roof-mounted
omni-directional
antenna
PC’s ethernet port
provides Internet
service to user
802.11b card based on
Intersil Prism 2.5 chipset
RTS/CTS disabled,
pseudo-IBSS mode
Software and
Auto-configuration






Each node: Linux, routing software implemented in
Click, a DHCP server, a web server
Software is pre-installed.
Node acts as like a cable or DSL modem
User connects PC or laptop to the node’s ethernet
interface
Node automatically configures user’s computer via
DHCP
Lists itself as default IP router
Addressing






Roofnet carries IP packets inside its own header
format and routing protocol
Node: chooses address whose low 24 bits are low
24 bits of node’s Ethernet address and high 8 bits
are an unused class-A IP address.
Same address at both the Roofnet and IP layers
These addresses are meaningful only inside
Roofnet
Allocates addresses from 192.168.1.x to users
NAT between Ethernet and Roofnet
Gateways and Internet Access





Each node on startup asks for an IP address
as a DHCP client.
If it succeeds, the node advertises itself as an
Internet gateway.
Gateway acts as NAT for connections from
Roofnet to the Internet.
Node selects gateway to which it has the best
route metric.
Four Internet gateways
Routing Protocol





Srcr - find highest throughput route between
pair of nodes
Omnidirectional antennas give choice of links
Dynamic source-routing (DSR)
Each node maintains partial database of link
metrics
Dijkstra’s algorithm
Routing Protocol

Link metric learning:



Node includes link’s current metric in packet’s
source route
DSR-style flooded query
Overheard queries and responses
Routing Protocol




Combination of link-state and DSR-style on
demand querying
Roofnet gateway floods dummy query
Node sends data to a gateway – gateway
learns about links back to the node
Nodes do not need to send flooded queries
Routing Protocol





Flooded queries often do not follow best
route
Srcr solution – compute best route from
database
Link conditions change leading to change in
best route
Notification of failed link sent back to source
New metric information sent to source
Routing Protocol


Source re-runs Dijkstra’s algorithm
Better metric information:


Sources learn through dummy queries from
gateways or
Unsolicited link metric information about nearby
links
Routing Metric



Srcr uses Estimated Transmission Time
(ETT) metric
ETT predicts total amount of time needed to
send data packet along a route
Srcr chooses route with lowest ETT
Routing Metric



“Srcr predicts that a link’s highest-throughout
bit-rate is the bit-rate with the highest product
of delivery probability and bit-rate.”
1500-byte periodic broadcasts at each
available 802.11 bit rate
Periodic minimum-size broadcasts at 1Mbps
Routing Metric


“ETT metric for a link is the expected time to
send a 1500 byte packet at that link’s highest
throughput bit-rate.”
ETT metric for a route is the sum of the ETTs
for the route’s links.
t = 1 / Σi 1/ti
t = route’s end-to-end throughput
ti = throughput of route’s hop

Bit-Rate Selection





SampleRate: Roofnet’s algorithm to choose among
802.11b transmit bit rates.
Adjusts bit-rate as it sends data packets over a link
Adjusts choice more accurately and quickly than
ETT
Bases choice on actual data transmission v/s on
periodic broadcast probes
Sends packets at bit-rate which currently provides
highest throughput
Evaluation







Method
Basic Performance
Link Quality and Distance
Effect of density
Mesh Robustness
Architectural Alternatives
Inter-hop Interference
Method

Multi-hop TCP data set





84-byte pings once per second for 10 seconds
Route established and latency measured
Throughput = number of bytes delivered to
receiving application
10% pairs: no working routes
Single-hop TCP data set

Measure throughput on direct radio link between
pair of nodes
Method

Loss matrix data set



Multi-hop density data set



Measure loss rate between pair of nodes
1500-byte broadcasts at each 802.11b bit-rate
Measure throughput between fixed set of four nodes
Vary number of nodes participating in routing
Some of the analyses involve simulated route
throughput calculated from the single-hop TCP.
Basic Performance
Average throughput is 627 kbits/sec
Basic Performance
TCP throughput to each node from its chosen gateway
Link Quality and Distance


Srcr favors short links
of a few hundred
meters.
Fast, short hops are the
best policy
Link Quality and Distance


Median = 0.8
Single-hop route with
40% loss can deliver
more data than a twohop route with perfect
links.
Effect of density



“Mesh networks are effective only if the node
density is sufficiently high.”
Simulate different size subsets of Roofnet
Estimate multi-hop throughput between pairs
in the subset
Effect of density
Mesh Robustness
Most nodes have many neighbors
Majority of nodes use many neighbors
Roofnet makes good use of the mesh architecture in ordinary routing
Mesh Robustness


Extent to which network
is vulnerable to loss of
its most valuable links
Dozens of the best links
must be eliminated
before throughput is
reduced by half.
Mesh Robustness


Effect on throughput of
cumulatively eliminating
the best-connected
nodes.
Best two nodes are
important for
performance.
Architectural Alternatives:
Optimal Choice




Comparison with
single-hop (access
point) network.
Single-hop: 5 gateways
to cover all nodes
Multi-hop forwarding
provides higher
average throughput
Sequence of short high
quality links
Architectural Alternatives:
Random Choice



If Roofnet were singlehop, 25 gateways
would be required to
cover all nodes.
Multi-hop routing
improves connectivity
and throughput.
Careful gateway choice
improves throughput for
both multi-hop and
single-hop routing.
Inter-hop Interference


Measured multi-hop
throughput lower than
expected.
Concurrent
transmissions on
different hops collide
and cause packet loss.
Inter-hop Interference


802.11 RTS/CTS
mechanism: prevent
collisions
RTS/CTS does not
improve performance
Network Use






Measurements of user activity on Roofnet
One of the four gateways monitored packets
forwarded between Roofnet and the Internet.
In one 24-hr period: Average of 160 kbits/sec
Gateway’s radio busy 70% of the monitoring
period
Less than 1% UDP. Rest were TCP.
16 Roofnet nodes accessed the Internet
Conclusion





Ease of deployment
37 nodes in one year with little administrative
or installation effort
Average throughput = 627 kbits/sec
Position of internet gateways determined by
convenience
Multi-hop mesh increases both connectivity
and throughput
Questions?