Transcript The Pulse Protocol - Department of Computer Science

High Performance
Mobile Ad hoc Networking
Herbert Rubens
Baruch Awerbuch
[email protected]
[email protected]
Johns Hopkins University
Department of Computer Science
Wireless Communication Lab
wireless.cs.jhu.edu
Presentation Overview






Mobile Ad hoc Networking Overview
Research Contributions
Related Work
The Pulse Protocol
The Medium Time Metric
Wave Relay System
Feel free to ask questions throughout the presentation!
Mobile Ad hoc Network





A self configuring network of mobile routers
connected by wireless links
The routers may move freely, creating arbitrary
network topologies
The network topology can change rapidly and
unpredictably
Nodes communicate by wirelessly forwarding or
relaying data through intermediate nodes
The network can be connected to the larger
Internet or operate independently
http://en.wikipedia.org/wiki/Mobile_ad-hoc_network
JHU Wave Relay Network
Node Locations Determine Topology
Mobile Ad hoc Networking Timeline
Destination Sequenced
Distance Vector (DSDV)
Functions and Structure of
a Packet Radio Station
CD
Player
Apple
Founded
1975
Microsoft
Founded
Herbert
Benjamin
Rubens
1979
Apple
IIgs
1985
DARPA Packet Radio
Networks
Intel
486
Windows
3.0
1991
www
Ad hoc On-demand
Distance Vector (AODV)
Y2K
1995
Dynamic Source
Routing (DSR)
Today
Optimized Link-State
Routing Protocol (OLSR)
Burchfiel, J., Tomlinson, R., Beeler, M. (1975). "Functions and structure of a packet radio station". AFIPS: 245.
Kahn, R. E. (January 1977). "The Organization of Computer Resources into a Packet Radio Network". IEEE Transactions on Communications COM-25 (1): 169–178.
Kahn, R. E., Gronemeyer, S. A., Burchfiel, J., Kunzelman, R. C. (November 1978). "Advances in Packet Radio Technology". Proceedings of IEEE 66 (11): 1468–1496.
Jubin, J., and Tornow, J. D. (January 1987). "The DARPA Packet Radio Network Protocols". Proceedings of the IEEE 75 (1).
Fundamental Challenges

Complex dynamics of a wireless link

Continuously fluctuating RF environment (without mobility!)

Bit Error Rate


Modulation




Different modulations work better in different RF environments
Multi-path, channel fading, delay spread
Link Capacity
Mobility


Further increases wireless link dynamics
Creates hard transitions


= small packets more reliable then large packets
walk around a corner and everything changes
If all of the links are continuously changing, how do you
select a set of links to form a path?
Research Objectives

Scalability



Design routing algorithms which scale to thousands of
devices while minimizing control overhead
Routing algorithm must perform under vehicular
mobility, urban channel fading, and arbitrary
communication patterns
Efficiency

Selected routes must:



Maximize individual path capacity
Minimize network resource consumption
Continuously adapt to changes
Research Contributions

Medium Time Metric (MTM)




Pulse Protocol




First route selection metric to consider multi-rate radios
Provably optimal route selection in small to medium sized
networks
Experimental results and simulated results validate approach
Extremely scalable routing protocol designed for mobile
networks
Optimized for infrastructure access and peer-to-peer traffic
patterns
Protocol extensions provide integrated time synchronization and
power saving
Sensor Network Pulse Protocol



Directly trades route activation delay for power saving efficiency
Optimized for infrequently changing sensor network topologies
Optimized for sensor to collector traffic model
Publications
Relevant to Thesis

MONET Journal – “The Medium Time
Other work

Algorithms for the On-Demand Secure
Byzantine Routing Protocol”
Metric: High Throughput Route Selection in
Multi-rate Wireless Networks”

WONS 2005 – “The Pulse Protocol:
ESAS 2006 – “Dynamics of Learning

SECURECOM 2005 – “On the
Survivability of Routing Protocols in Ad Hoc
Wireless Networks”
Mobile Ad hoc Network Performance
Evaluation”

MILCOM 2004 – “The Pulse Protocol:

NDSS 2005 – “Secure Multi-hop

INFOCOM 2004 – “The Pulse Protocol:

INFOCOM 2005 – “Provably Competitive
WONS 2004 – “High Throughput Route

IZS 2004 – “Swarm Intelligence Routing

WiSE 2002 – “An On-Demand Secure

Sensor Network Routing and Power Saving”
Energy Efficient Infrastructure Access”
Selection in Multi-rate Wireless Networks”
Infrastructure Access”
Adaptive Routing”
Resilient to Byzantine Adversaries”
Routing Protocol Resilient to Byzantine
Failures”
Existing Approaches
Receivers
Urban Channel Environment
10
5
• Multi-path fading & shadowing
0
-5
• Rapidly changing channel conditions
-10
-15
-20
0
200
400
600
800
1000
1200
1400
1600
1800
2000
Reactive On-Demand Protocols
(AODV, DSR)
10
5
• On-demand protocols have no prior
knowledge of channels conditions
0
-5
-10
• A RREQ packet provides only a single
sample of a complex distribution
-15
-20
0
200
400
600
800
1000
1200
1400
1600
1800
2000
10
5
0
-5
-10
Destination
-15
Source
-20
-25
• Channel is continuously changing
-30
-35
-40
Proactive Link State Protocols
(OLSR, TBRPF)
0
200
400
600
800
1000
1200
1400
1600
1800
2000
• Continuous flooding from every node in
the network
10
5
0
-5
• Hello Protocol – detects link changes
-10
-15
-20
-25
-30
-35
-40
0
200
400
600
800
1000
1200
1400
1600
1800
2000
You can not accurately track channel
with control packets!
How Often Does Connectivity Change?







10% of min-hop paths fail within 1.3
seconds
After 5 seconds 25% of min-hop
paths have failed
On-Demand routes may only work
for a short period of time
Link State Protocols need to flood
every time a link changes
These simulations only consider
changes from connected  not
connected (in free space)
What about changes in link speed?
Reliability? Hard transitions in a real
environment? Fast-fading and urban
channel effects?
Connectivity is continuously
changing at an extremely fast rate!
Simulation:
• 100 Nodes
• 1000m x 1000m area
• Random Waypoint Mobility (Max
Speed=20m/s)
• Calculate All-to-All shortest path initially,
then track how long until the route fails
Pulse Protocol Outline

Pulse Protocol Overview



Scalable multi-hop ad hoc routing protocol
Based on Tree Routing
Tree Routing vs. Direct Routing
The Pulse Protocol

Proactive Component





Tracks minimum amount of information to avoid flooding for
route establishment and maintenance
Periodic flood operation (similar to Hello Protocol)
Proactively rebuilds spanning tree
Estimates neighbors, density, SNR, loss rates, capabilities,
number of radios, MTM metric
On-Demand Component

Route establishment


Using only UNICASTS!
Gratuitous mechanism




Neighbors promiscuously monitor packets
Metric tracked at the speed of data packets NOT control packets!
Path switches as metrics change
Local changes in connectivity only generate local traffic

Unlike BOTH on-demand and link state protocols
Ad hoc Nodes
Network Connectivity
Pulse Flood
Spanning Tree
Source and Destination Need to
Establish a Path
Pulse Response Sent to Root
Destination Paged on Next Pulse
Destination Sends Pulse Response
Path Option 1: Through the Root
Through the Root Path
9 Hops
Shortest Path
2 Hops
This option is inefficient! It is not necessary to go to the root. Better routes already exist!
Path Option 2: Tree Traversal
Tree Traversal Path
5 Hops
Shortest Path
2 Hops
Path Option 3: Tree Shortcut
Tree Shortcut Path
3 Hops
Shortest Path
This is the initially selected path of the Pulse protocol.
2 Hops
Path Optimization: Gratuitous Reply
Selected Path
2 Hops
Shortest Path
2 Hops
Node sends gratuitous reply
Proactive Route Maintenance
Proactive Route Maintenance
Tree Routing vs. Direct Routing

Direct Routing

Attempts to initially discover the shortest path


Link state



tracks every link in the network regardless of whether it is used
a shortest path spanning tree for every node in the network
On-Demand




Requires large overhead
floods the network to establish a route
re-floods when ever the path breaks
a shortest path spanning tree for all nodes transferring data
Tree Routing




Proactively rebuilds a single spanning tree on top of the network
Boot straps communication off of the tree route
Route are not initially the direct shortest path, but routing mechanism
allows the path to converge towards the shortest path
Active destinations can be reached without flooding the network

Efficient operation for realistic traffic patterns
Pulse Protocol Concepts

Aggregation – for scalability



Spanning tree represents a compressed view of the network
topology
Pro-active component maintains the minimum amount of
information to allow efficient route establishment
De-Aggregation – for efficiency




The routing metric is tracked at the speed of the data flow
Changes to the metric are only reported locally
Routes are continuously adjusted as the metrics change
High speed accurate route tracking is essentially an on-demand
decompression of the topology


However, it occurs ONLY in areas of the network with active data
flows
Result: a scalable routing structure which tracks paths at
the speed of the data flow
Internet Gateway Example
• All nodes routing to centrally
located internet gateway
• Best possible case for Pulse
Protocol
• Pulse source is designated
as the centrally located
gateway
• Representative of Pulse
internet access deployment at
JHU
• Similar to DoD “Reach
Back” model
Representative of most common
real-world communication model
Delivery Ratio Simulations
DSR
Pulse
• Pure peer-to-peer
communication pattern
• Pulse source is an
arbitrary mobile node
SNS Scalability Simulation
10 km





Size: 10 km x 10 km
Nodes: 5,000
Speed: 1 m/s
Traffic: 5 Mbps
Delivery Ratio: 97.2%
10 km
Links: 50,000 on average
• 100 stationary backbone nodes were arranged in a 10 by 10 grid
• 5000 nodes were randomly placed and moved randomly
• Exponential random traffic pattern was used
• A network of 5,000 nodes could contain up to 25 million wireless links.
Medium Time Metric Outline



Why do wireless radios operate at
multiple rates?
Minimum Hop Metric shortcomings
Medium Time Metric
Advantage of Multi-Rate?
1 Mbps

2 Mbps
5.5 Mbps
11 Mbps



• 802.11g
• 1,2,5,6,11,12,18,24,36,48,54
Mbps
• 802.11n (draft)
• A lot more! Up to 300 Mbps.
Direct relationship between
communication rate and
the channel quality
required for that rate
As distance increases,
channel quality decreases
Therefore: tradeoff
between communication
range and link speed
Multi-rate provides
flexibility
Challenge to the Routing Protocol



Must select a path from Source to
Destination
Links operate at different speeds
Fundamental Tradeoff


Fast/Short links = low range = many
hops/transmissions to get to destination
Slow/Long links = long range = few
hops/transmissions
Minimum Hop Metric
(Traditional Technique)




Not designed for multi-rate networks
A small number of long slow hops provide
the minimum hop path
These slow transmissions occupy the
medium for long times, blocking adjacent
senders
Selecting nodes on the fringe of the
communication range results in reduced
reliability
New Approach:
Medium Time Metric (MTM)


Assigns a weight to each link proportional
to the amount of medium time consumed
by transmitting a packet on the link
Enables the Pulse protocol to discover the
path that minimizes total transmission time
MTM Example
Medium Time Usage
Destination
Link Throughput
11 Mbps
2.5ms
4.55 Mbps
5.5 Mbps
3.7ms
3.17 Mbps
2 Mbps
7.6ms
1.54 Mbps
1 Mbps
13.9ms
0.85 Mbps
Source
Path Medium Time Metric (MTM)
Path Throughput
11 Mbps
5.5 Mbps
2 Mbps
1 Mbps
1
13.9ms
= 13.9 ms
0.85 Mbps
MTM Example
Medium Time Usage
Destination
Link Throughput
11 Mbps
2.5ms
4.55 Mbps
5.5 Mbps
3.7ms
3.17 Mbps
2 Mbps
7.6ms
1.54 Mbps
1 Mbps
13.9ms
0.85 Mbps
Source
Path Medium Time Metric (MTM)
11 Mbps
5.5 Mbps
2 Mbps
1 Mbps
5.5 + 2
1
3.7ms
13.9ms
7.6ms
Path Throughput
= 11.3 ms
= 13.9 ms
1.04 Mbps
0.85 Mbps
MTM Example
Medium Time Usage
Destination
Link Throughput
11 Mbps
2.5ms
4.55 Mbps
5.5 Mbps
3.7ms
3.17 Mbps
2 Mbps
7.6ms
1.54 Mbps
1 Mbps
13.9ms
0.85 Mbps
Source
Path Medium Time Metric (MTM)
11 Mbps
5.5 Mbps
2 Mbps
1 Mbps
11 + 2
2.5ms 7.6ms
5.5 + 2
3.7ms
1
13.9ms
7.6ms
Path Throughput
1.15 Mbps
= 10.1 ms
= 11.3 ms
= 13.9 ms
1.04 Mbps
0.85 Mbps
MTM Example
Medium Time Usage
Destination
Link Throughput
11 Mbps
2.5ms
4.55 Mbps
5.5 Mbps
3.7ms
3.17 Mbps
2 Mbps
7.6ms
1.54 Mbps
1 Mbps
13.9ms
0.85 Mbps
Source
Path Medium Time Metric (MTM)
11 + 11
11 Mbps
5.5 Mbps
2 Mbps
1 Mbps
2.5ms 2.5ms = 5.0 ms
11 + 2
2.5ms 7.6ms
5.5 + 2
3.7ms
1
Path Throughput
13.9ms
7.6ms
2.38 Mbps
1.15 Mbps
= 10.1 ms
= 11.3 ms
= 13.9 ms
1.04 Mbps
0.85 Mbps
MTM Advantages

Paths which minimize network utilization,
maximize network capacity



Avoiding low speed links inherently provides
increased route stability


Global optimum under complete interference
Excellent heuristic in even larger networks
High speed links operate with greater margin and are
more elastic under changes
Experimental results show up to 17 times
greater throughput using MTM in 802.11g
networks
Wave Relay System
and Test-bed
Wave Relay Test-bed

Over 50 Wave Relay Routers deployed
across JHU Campus





Urban City Environment
Internet Access, Ad hoc Access Points,
Voice over IP
Mobility testing from automobiles
Over 100 JHU students simultaneously
use network each day for Internet Access
System tested at Holcim Industrial Plant
(Chicago, IL)

Complex propagation environment

Enabled real-time industrial process
control

Currently Deployed Custom Applications

Military Distributed Battlefield Mapping



GPS based interactive map
Eventual reliability
Locality Specific Messaging System


GPS based messaging system
Messages targeted to any user at a
specific location
Wave Relay Device
Software

Pulse Protocol [Infocom’04, Milcom’04, WONS’05]

Scalable ad hoc routing protocol

Active path tracking

Based on Tree Routing strategy
Medium Time Metric [MONET,WONS’04]

High Throughput Path Selection

Increased Path Elasticity

Efficient Multi-rate Operation
Hardware




Handles merge, partition, failure
Embedded Linux Distribution










Leader Election Algorithm


Embedded Single Board Computer
Less then 8 MB storage requirement
Linux Kernel Module 2.4 and 2.6 compatibility


Operates at layer 2
Distributed virtual switch architecture provides
seamless bridging
Intel IXP425 Network Processor
On-chip Cryptographic Accelerator
64 Mb Ram onboard
4 mini-PCI interfaces
Dual 10/100 Ethernet
Compact flash interface
Serial port / JTAG / GPIO
Hardware Watchdog
Power over Ethernet

+9V to +48V DC Input

Atheros 802.11g/b Wireless Card

16 MB Intel Strata Flash


Garmin GPS 16 receiver
Li-Ion Battery Pack

Industrial NEMA 67 Enclosure







400 mW (26 dBm) output power
Stores OS & Wave Relay software
~20 hours continuous runtime
4 N-type antenna mounts
2 Ethernet Ports
(6) protection against dust
(7) water submersible
Wireless Shuttle Bus Project
http://wireless.cs.jhu.edu/mobile/
Conclusion

The Pulse Protocol provides scalable routing
under high levels of mobility

The Medium Time Metric selects high
throughput routes and minimizes consumption of
the shared wireless medium

The Wave Relay test-bed demonstrates the
effectiveness of the Pulse + MTM protocols in a
real-world urban environment
Thank You!
Questions?