Context-aware Adaptive Routing for Delay
Tolerant Networking
Mirco Musolesi
Joint work with Cecilia Mascolo
Department of Computer Science
University College London
Delay Tolerant Networks Research Group Meeting
65th IETF Meeting - Dallas, 23 March 2006
Outline of this Talk
• Overview of the Context-aware Adaptive Routing (CAR)
protocol
• Extension to hybrid networks
• Integration with the DTN reference implementation
• Current research directions
Context-aware Adaptive Routing (CAR)
• Routing protocol for delay tolerant networks initially
designed for delay tolerant mobile ad hoc networks
• Extension to hybrid networks, where some nodes are fixed
or act as gateways for interconnecting different regions
• Context-aware routing designed for scenarios where:
– Deterministic routing information is not available
– No geographical information (like GPS coordinates) are available
Adaptive Routing for Intermittently Connected Mobile Ad Hoc Networks
Mirco Musolesi, Stephen Hailes and Cecilia Mascolo
In Proceedings of 6th IEEE International Symposium on a World of Wireless, Mobile and Multimedia
Networks (WoWMoM'05). Taormina, Italy. June 2005.
Overview of the Protocol
• Based on host(s) acting as carrier(s) for asynchronous
delivery of messages to hosts that can be:
– Final recipient of the message/bundle
– Gateways between regions (fixed or mobile, such as mobile sinks)
• Choice of the best carrier(s) based on the evaluation of the
context information
–
–
–
–
Host colocation
Host mobility
Battery level
…
The Novelty of Our Approach
• Novelty of our approach: store-and-forward decisions
based on the prediction of the evolution of the DTN
scenario
• In particular, we use Time Series Analysis based on State
Space Models (Kalman Filter) to keep history into account
and to predict the evolution of DTN scenarios
Asynchronous Delivery: An Example
C
C
B
P
C
A
M
R
D
E
N
F
I
H
L
Q
O
Calculation of the Host Utility
• Host utility calculated using multi-criteria decision theory
– A utility is associated to each context attribute (i.e., a utility
associated to host colocation
– Utilities are then composed using a weighted functions
Calculation of the Host Utility
• We tested the algorithm considering two attributes
– Colocation with a certain host (that may be the final recipient of the
message or a gateway)
– Change degree of connectivity
Kalman Filter Prediction
• Very lightweight
– Only information about the current state must be maintained
– Suitable for resource-constrained devices
• No “learning” phase is necessary
– Fast convergence of the filter
• Different prediction models (considering trends and
seasonal/periodic behavior)
Context-aware Adaptive Routing Protocol
• Host utilities are calculated by each host and are sent to
the others together with the routing tables (in our
implementation DSDV)
– Based on local calculations
– Adaptive refresh interval based on the context variability
• If the carrier of the message gets in reach with a host that
can guarantee a better delivery probability, the message is
transferred to that host
Predictability of Context Information
• Key issue for CAR: predictability of context information
– Forecasting model cannot provide accurate predictions due to the
inherent characteristics of the time series (for example
randomness)
• Design of an autonomic component that is able at runtime
to analyze the predictability of the time series of the
context information
• If context information is not predictable, alternative routing
strategies can be adopted:
– Intelligent epidemic routing
– …
Evaluating Context Information Predictability for Autonomic Communication
Mirco Musolesi and Cecilia Mascolo
In Proceedings of 2nd IEEE Workshop on Autonomic Communications and Computing (ACC'06).
Niagara Falls, NY. June 2006. To appear.
Integration of CAR in the DTN Reference
Implementation
• Implementation in the DTN2 reference implementation in
process (testing phase)
• Porting to resource-constrained devices
• Integration of DTN reference implementation in a hybrid
environment
– DTN reference implementation for gateways?
– Integration with other DTN systems for constrained devices
(Bluetooth equipped mobile phones, sensor networks).
CAR Implementation in DTN2
QuickTime™ and a
TIFF (Uncompressed) decompressor
are needed to see this picture.
CAR for Multi-Regions DTNs
• Based on deterministic and probabilistic mechanisms
– For example a server connected to the Internet has delivery probability
equal to 1 for the smtp domain.
– Message ferries will have a delivery probability equal to 1 to hosts that are
on their pre-defined paths
• Host utilities used to measure
– The probability of delivering message to the recipients of the message for
intra-region communication
– the probability of delivering messages to gateways (or sinks in
sensor networks) for inter-regions communication
• Deployed on heterogeneous devices: mobile phones and
WiFi APs (in progress), sensor networks (planned)
Our Current Research Directions
• Implementation of a system based on CAR for interregions communications
– CAR for sensor networks with mobile sinks
• Porting of DTN for Contiki (SICS, Sweden)
• Content based routing based on CAR
• Design of realistic mobility models
– based on social networks theory
– validated using real traces
SCAR: Context-aware Adaptive Routing in Delay Tolerant Mobile Sensor Networks
Cecilia Mascolo and Mirco Musolesi
CS-UCL Research Note. January 2006.Submitted for Publication.
Questions?
Adaptive Routing for Intermittently Connected Mobile Ad Hoc Networks
Mirco Musolesi, Stephen Hailes and Cecilia Mascolo
In Proceedings of 6th IEEE International Symposium on a World of Wireless, Mobile and Multimedia
Networks (WoWMoM'05). Taormina, Italy. June 2005
Evaluating Context Information Predictability for Autonomic Communication
Mirco Musolesi and Cecilia Mascolo
In Proceedings of 2nd IEEE Workshop on Autonomic Communications and Computing (ACC'06). Niagara
Falls, NY. June 2006. To appear.
SCAR: Context-aware Adaptive Routing in Delay Tolerant Mobile Sensor Networks
Cecilia Mascolo and Mirco Musolesi
CS-UCL Research Note. January 2006.Submitted for Publication.
Mirco Musolesi
Department of Computer Science University College London
[email protected] http://www.cs.ucl.ac.uk/staff/m.musolesi