Rumor Routing Algorithm For Rumor Routing Algorithm For sensor Networkssensor Networks

David Braginsky, Computer Science Department, UCLA

Presented By: Yaohua Zhu

CS691 Spring 2003

OutlineOutline Introduction Flooding Event Flooding Query Rumor Routing Algorithm Agents Query Simulation Results Related Work Future Work

Wireless Sensor NetworkWireless Sensor Network

Wireless communication capability– Emerging low power– Small form-factor processors

Sensor– Allow for larger-scale, extremely dense

network

Design ConsiderationDesign Consideration Each node does not posses significant computational

power Sensing highly distributed Algorithms highly distributed, because local

communication with stringent power requirements Self-configuring, Highly scalable, redundant Robust with shifting topologies Gather data from different parts of network Without taxing its limited bandwidth and power Reduce failure rates

Basic IdeaBasic Idea

Routing queries to nodes that have observed a particular event

Retrieve data on the event

Event and QueryEvent and Query Event

– An abstraction, identifying anything from sensor readings

– Assumed be localized phenomenon – Occurring in a fixed region of space

Query– Be request for information– Orders to collect more data– Query arrives destination, begin to flow back to query’s

originator

Flooding Event and Flooding Flooding Event and Flooding QueryQuery

Flood event– For few events and many queries– Set up gradients towards it

Flood query– For less queries per event– Less data generated by each event

Query FloodingQuery Flooding Assume no collisions For N nodes we must perform N transmissions per

queries For Q queries the transmissions total is

N * Q Energy used is independent of the number of

events tracked by network Useful when the number of events is very high,

compared to the number of queries

Event FloodingEvent Flooding When node witnesses an event, it can flood the

network. All other nodes form gradients toward the event

N is transmissions per event E is the number of events The total energy expended by event flooding is

E * N It is independent of the number of queries Efficient when the number of events is low, compared

to the number of queries

Query Flooding and Event Query Flooding and Event FloodingFlooding

Idea of Rumor Routing Idea of Rumor Routing AlgorithmAlgorithm

Fill the region between query flooding and event flooding

Only useful if the number of queries compared to the number of events is between the two intersection points

An application of this ratio can use a hybrid of rumor routing and flooding to best utilize available power

Algorithm OverviewAlgorithm Overview Assume network consists of densely distributed

wireless sensor nodes with relatively short symmetric radio range

Nodes records events and able to route queries Each node maintains a list of neighbors, events table,

forwarding information to all the events it knows. Neighbors list created and maintained by actively

broadcasting a request Since the simulation were static topology, each node

broadcast its id at the beginning

Contd…(node)Contd…(node)

When node witnesses an event, it adds it to its event table, with a distance of zero to the event

Node also has a random chance to generating an agent

The probability of generating an agent is an algorithm parameter

Contd…(Agent)Contd…(Agent)An agent is a long-lived packetAgent travels the networkPropagating information about local events

to distant nodesContains events tableSynchronizes with every node it visitsTravels network some number of hops, then

dies

Contd…(query)Contd…(query) Any nodes may generate a query and routed to a

particular event If node has a route to the event, it will transmit the

query It it does not, it will forward the query in a random

direction Continues until the query TTL expired, or query

reaches a node that has observed the target event If the node originated the query did not reach a

destination, it can always flood the query

AgentsAgents

Each agent informs nodes it encounters of any events along its route

Carries a list of eventsAlong with the number of hops to that eventWhen it arrives node A from neighbor B, it

synchronize its list with the nodes list

Contd…(Agents)Contd…(Agents)

A’s route to E1 is longer than the agent’s

Agent does not know to route to E2

Contd…(Agents)Contd…(Agents)

After the table synchronization completes, the event table will contain the best routs to the event

Contd…(Agents)Contd…(Agents)

Straightening algorithm used to determine the agent’s next hop

Agent maintains a list of recently seen nodes When arrives a node, it adds all node’s neighbors

to the list When picking next hop, it first try nodes not in the

list It allows agent to create fairly straight paths

Contd…(Agents)Contd…(Agents)

Policy to generate agentNode that witnessed an event generate an

agentThe number of agents depends on the

number of event, event size, and the node density

Event table have expiration timestamp

QueriesQueries A query can be generated at any time by any nodes and

target to an event If a node has a route toward target event, it forwards the

query along the route If it does not, forward to random neighbor and assume the

query has not exceeded its TTL Query employs same mechanism as the agent, keep list of

recently seen nodes Some queries may not reach their destination, application

must detect the failure, flooding query again or increase the queries TTL

Rumor RoutingRumor Routing Create paths leading to each event Event flooding creates a network-wide gradient

field Query sent random walk until find the event path No flooding event across the network Query discovers event path, then route directly to

the event If path cannot be found, application re-submitting

the query, flooding it

Set up PathSet up Path

Contd…Contd…

Agent A1 create path state leading to E1

Agent A2 create path state leading to E2

When A2 crosses the path created by A1, it create aggregate path state leading to E1 and E2

Contd…Contd…

Agent optimize the path if they find shorter ones When agent find a node route to event is more

costly than its own, it will update the node’s routing table to efficient path

Comparison Event Routing Comparison Event Routing and Query Routingand Query Routing

Query flooding: Et = Q * NEvent flooding: Et = E * NThe algorithm has addition energy for path

setup and query routingEt = Es + Q * (Eq + N * (1000 – Qf) / 1000 )– N * (1000 – Qf) additional send– Qf is the number of delivered queries– Eq is the energy spent routing queries

Simulation ResultsSimulation Results

Simulation ResultsSimulation Results

Algorithm StabilityAlgorithm Stability

Because this algorithm relies on random decision(agent and queries)

Performance not vary significantly over several runs is important

To test same set of parameters run 50 simulations, Average Te 118, 99% of the values for Te will be found between 104 and 131

Algorithm is stable for particular configuration

Fault ToleranceFault Tolerance

After the routes were established some of the nodes were disabled

Probability of delivery degraded slowly for 0-20% node failure

Over 20% node failure, the performance degraded more severely

Related WorkRelated Work

GRAdient BroadcastGossip RoutingAnt AlgorithmDirected Diffusion and Geo-Routing Data-Centric Storage in Sensornets

GRAdient Broadcast(GRAB)GRAdient Broadcast(GRAB) Build a cost field toward a particular node, then

reliably routing queries across a limited size mesh toward that node

With overhead of network flood Queries route along short paths Delivered cheaply and reliably Not designed specifically to support the network

process, influenced the work of event-centric routing state in network

Gossip RoutingGossip RoutingNodes flood by sending message to some of

neighborsBy the redundancy in the link, most nodes

receive the flooded packedUsed to deliver query or flood eventsLess overhead than conventional floodingNot be designed specifically for energy

constrained contexts

Ant AlgorithmAnt Algorithm Agent traverse the network encoding the quality of the

path they have traveled, and leave it the encoded path as state in the nodes

At every node, an agent picks its next hop probabilistically, biased toward already know good paths

Faster and more through exploration good regions Very effective in dealing with failure, because always

some amount if exploration But due to large number of nodes, the ant agents required

to achieve good results

Directed Diffusion and Geo-Directed Diffusion and Geo-RoutingRouting

Provide a mechanism for doing a limited flood of a query toward the event

Set reverse gradients to send data back along the best route

Results in high quality pathsBut requires an initial flood of the query for

exploration

Data-Centric Storage in Data-Centric Storage in SensornetsSensornets

Allow access to named data by hashing the name to a geographic region in the network

Used to efficiently deliver queries to named events by storing the location of the events

Relies on a global coordinate system

Rumor Routing AlgorithmRumor Routing Algorithm Presents good alternative to event and query

flooding Performance depend on event distribution, guarantee

better results than event flooding Successful under all simulated node and event

densities Difficult to predict the max query to event ratio No reliable trend has been found in the max query to

event ratio with increasing node and event densities

Summary SlideSummary Slide

Future WorkConclusionReferences

Future WorkFuture Work Wider range of simulation scenarios

– Network dynamics– Consider collisions– Asynchronous event– Non localized event– Non random query pattern

Algorithm design alternatives– Non random next hop selection– Use of constrained flooding– Parameter setting exploration