Probabilistic routing in intermittently connected networks

Probabilistic Routing in Intermittently Connected Networks Anders Lindgren, Avri Doria, Olov Schelen
Presenter: Mihai Fonoage (mfonoage@)
Wireless and Sensor Networking (WISENET) Group
Outline •Introduction •Background
•Related Work •Probabilistic Routing •Simulations
•Results
•Discussion and Future Work •Conclusions •References
Introduction
•In ad hoc networks nodes are equipped
with low power radios
•Communication Ùfully connected path
•This is not the case in many scenarios such
as in Intermittently Connected Networks
•One solution:
•Buffer messages at intermediate nodes +
Exploit mobility of those nodes +
Transfer messages to other nodes as they meet
Introduction (cont.)
Background •Intermittently Connected Networks= fully connected path between source and destination might not exists all the time
•Systems where such a communication network has been deployed:
•DakNet Project: store-and-forward networks
connecting villages through relays on buses and
motorcycles in India and Cambodia
•Military war-time scenarios and disaster recovery
situations: soldiers in hostile environment => no
infrastructure can be assumed
•Sensor networks: attaching sensors to seals or whales,
thus increasing the number of oceanic temperature
readings. Sink would be deployed in feeding areas
Background (cont.)•ZebraNet: Zebras are equipped with tracking collars
•Weather monitoring in large areas: boards show the weather reports from other parts of the park –done by equipping hikers with small network devices
Related Work
Epidemic Routing
•Relies on the theory of epidemic algorithms •Pare-wise information of messages between nodes
•Hosts buffer messages even if no path to the destination is available
•As nods get in contact they exchange summary vectors(index of buffered msg.)
•Determine if there are messages unseen
•Request these messages from the other node
Related Work
Epidemic Routing (cont.)
•As long as buffer space is available, messages will spread through the network as nodes meet
•Each message contains
•Globally unique message ID
•Used to determine if the message has been seen before •Source and Destination
•Hop Count –determines the maximum no. of hops a message can be sent (similar to the TTL field in IP packets)
Related Work
Other Work
•Shared Wireless Infostation Model (SWIM)•Nodes cooperate in transmitting information
form the network to Infostations
•Nodes give messages to other nodes based on a configurable probability
•Smart-tag based data dissemination
•Mobile smart-tags disseminate data gathered
from fixed nodes
•Pollen Network
•Allows for the possibility to have a centralized
unit called a hive used by mobile nodes to
synchronize their data
Related Work
Other Work (cont.)•Disconnected Transitive Communication (DTC)
•Utilizes a utility function to locate the node that is most suitable to forward the message
•Search for the node that is “closer”to the
destination
Probabilistic Routing
•Make use of the predictability of user movement
•Based on repeating behavioral patterns
•Ex: if a node has visited a location several times
before, it’s likely that it will visit that location again •Alternative to Epidemic Routing
•If buffer space and bandwidth are infinite,
Epidemic Routing gives an optimal solution
•Problem: buffer and bandwidth are scarce resources
most of the time
Probabilistic Routing
PROPHET •PROPHET = a Probabilistic Routing Protocol using History of Encounters and Transitivity
•Probabilistic metric called delivery predictability P(a,b)Є[0,1] at every node a for each destination node b.
•How likely a node is to deliver a message to
that destination
•Two nodes meet => exchange summary vector + delivery predictability vector
Probabilistic Routing
PROPHET (cont.)
•Delivery predictability calculation
•Update metric when a node is encountered
•Nodes often encountered have a high delivery predictability
•P(a,b) =P(a,b)old + (1 -P(a,b)old) x P init
•Aging is employed if two nodes do not encounter each
other in a while
•P(a,b) =P(a,b)old + γk
•γis the aging constant
•k is the no. of times units that have lapsed since the last time the
metric was aged
•Transitivity property that affects the delivery
predictability
•P(a,c) =P(a,c)old + (1 -P(a,c)old) x P(a,b) x P(b,c) x β
•βis a scaling constant that decides how large impact the
transitivity should have on the delivery predictability
Probabilistic Routing
PROPHET (cont.)•Forwarding strategies
•When a message arrives at a node there might
not be a path to the destination available
⇒Node has to buffer the message + upon each
encounter, decide whether to transfer the message •May also forward the message to multiple
nodes
⇒(+) Increase probability of message delivery
⇒(–) Increase resource consumption
•Chosen strategy: “when two nodes meet, a
message is sent to the other node if the delivery predictability of the destination of the message
is higher at the other node.”
Probabilistic Routing PROPHET Example
Simulations
Mobility Model
•Model mobility different than the random waypoint model to reflect reality
•“Community Model”shown below
•11 communities (C1-C11) and 1 gathering place (G)•Mobility scenario: nodes select a destination and a speed, move there, pause there for a while, and select a new
destination and speed
Destination selection
Simulations
Simulation Setup
•Metrics chosen:
• 1. Message delivery ability
•i.e. how many of the initiated messages is the protocol able to
deliver to the destination
• 2. Message delivery delay
•How much time it takes a message to be delivered • 3. Number of message exchanges
•Indicates how the system resource (such as bandwidth and
energy) utilization is affected by the different settings
Values for parameters kept fixed during simulation
Results
•Averages from 5 simulation runs
•Error bars in the graphs represent the 95% confidence intervals
•For each metric and scenario there are two groups with two different values of the hop count settings
•Each graph contains curves for both Epidemic Routing and PROPHET for the two different communication ranges •Different metrics versus queue size in the nodes
Results (cont.)
•Delivery rates of the protocols in the different scenarios
Results (cont.)
•Delivery rates of the protocols in the different scenarios
Results (cont.)•Delivery delay
Results (cont.)•Delivery delay
Results (cont.)•Communication overhead
Results (cont.)•Communication overhead
Discussion and Future Work •The paper describes an alternative to epidemically flooding messages through the network
•Increasing scalability
•The protocol uses a FIFO queue at the nodes
•Whenever a new message arrives to a full
queue, the message that has been in the queue
for the longest time is dropped.
•Other strategies can be used, like dropping the
message that has already been forwarded to the
largest number of nodes
Discussion and Future Work (cont.)•Make use of an ACK to reduce the required buffer space and to improve performance
•Messages already delivered can be removed
from the network
•=> More resources for other messages •Investigate other forwarding strategies •Limit the number of nodes that a message is
forwarded to => find the optimal number of
forwards to avoid performance degradation
Conclusions
•The authors propose the use of probabilistic routing
•Make use of non-randomness in node mobility •Defined a delivery predictability metric
•Reflects the history of node encounters, and transitive
and time dependent properties (aging)•PROPHET –uses the mentioned metric •Has better performance than Epidemic Routing in the
non-random scenario
•Has comparable performance with Epidemic Routing
in the random scenario
References
•Anders Lindgren, Avri Doria, Olov Schelen, “Probabilistic Routing in Intermittently Connected Networks”•Allan Beaufour, Martin Leopold, Philippe Bonnet, “Smart-Tag Based Data Dissemination”。