Autonomous Virtual Mobile Nodes Shlomi Dolev Seth Gilbert Elad Schiller Alex Shvartsman Jennifer Welch.

Autonomous Virtual Mobile Nodes

Shlomi Dolev Seth Gilbert

Elad Schiller Alex Shvartsman

Jennifer Welch

Challenges• Locality • Nodes only send messages to

nearby nodes

• Global coordination is expensive

• Locality • Unreliable nodes

• Mobile nodes fail,

• go to sleep, and

• get turned off.

Challenges• Locality • Unreliable nodes• Irregular motion

• Nodes travel wherever they want to go

• Locality • Unreliable nodes

• Mobile nodes fail,

• go to sleep, and

• get turned off.

Opportunities• Broadcast • Wireless broadcast is a powerful

primitive.

• Allows a node to reach all nearby

nodes,

• Ensure they receive the same

messages.

Opportunities• Broadcast• Time & Geography

• Nodes are physical entities– physical time & location

• Use GPS and/or algorithms for synchronization / location

Related work• Existing protocols • Flooding

• Distributed structure – E.g., TORA [PC97]

• Compulsory movement of nodes– [HP99, CNS01]

Related work• Existing protocols

• Random walk

• Random walk of a single agent

• Coping with chaos by chaos

Related work• Existing protocols

• Random walk

• Virtual nodes

• Geo-Quorums [DGL03]

• Virtual Stationary Automata– [DGL05]

• Virtual Mobile Node [DGL04]

Autonomous Virtual Mobile Node• Automaton • New programming abstraction

• A virtual general-purpose

computing entity.

Autonomous Virtual Mobile Node• Automaton • New programming abstraction

• Distinct location at any time

• Implemented by “real’’ mobile

nodes that happens to be near.

Autonomous Virtual Mobile Node• Automaton • New programming abstraction

• Distinct location at any time

• Communicates with:

• other virtual nodes, and

• “real” mobile nodes.

Autonomous Virtual Mobile Node• Automaton• Reliability

• Fault recovery

• The group emulation enhances

robustness:

• some may fail, or

• move out of range.

• Automaton• Reliability

• Fault recovery

• Self-stabilization

• Tolerate any starting state:

• maybe several (undesired)

copies, or

• none at all.

• Automaton• Reliability• Autonomous

• On-line movement decision: – current state, and

– sensor/environment input.

Example 1

• If north-east area appears deserted

• go south-west

Autonomous Virtual Mobile Node• Automaton• Reliability• Autonomous

• On-line movement decision: – current state, and

– sensor/environment input.

Example 2

• Hitchhike with the traffic, or

• Go in the opposite direction

Application Domain• Vehicular networks • Traffic control and safety

– E.g., ad hoc traffic light

Application Domain• Vehicular networks • Traffic control and safety

– E.g., ad hoc traffic light

Application Domain• Vehicular networks• RFID tags

• Very small, cheap and wireless tagging network.

• Limited power supply.– Photoelectric gate

• Use flash light to activate the net

• The AVMN follows the light

• E.g.,

• count the number of items

• find an expired item

• Use microwave instead of light.

Application Domain• Vehicular networks• RFID tags• Swarm computing

• Multiple virtual nodes

– Hierarchically originated

– Performing different task

• Collaborating or competing

Implementation• Exactly 1 instance Three different schemes

1. Virtual Stationary Automaton

• alive messages to known location

of a stationary node (VSA)

• VSA keeps track of the AVMN

• No message for too long

• create a new AVMN

• VSA eliminates duplicates

Implementation• Exactly 1 instance Three different schemes

1. Virtual Stationary Automaton

2. Send alive messages

• Send alive messages in a

random walk fashion.

• If a real node doesn’t receive an

alive message for too long

• generates a formation token

• carries ids and traverses

in a random walk fashion

• If tokens collide: merge ids’ lists

If containing more than (N+1)/2

• creates a new AVMN

Implementation• Exactly 1 instance Three different schemes

1. Virtual Stationary Automaton

2. Send alive messages

3. Nodes alive messages

• Real nodes periodically send

stay alive messages

• random walk to AVMN

• in order to survive

• AVMN must collect at least

(N + 1)/2 messages.

Implementation• Exactly 1 instance• Self-stabilization

• Every emulating real node– Keeps a replica of the AVMN

– Ensures identical replica

• Buffer input events waiting to be

applied to the state.

• At a fixed interval,

• sends replica to all.

• Predetermined function resolves

conflicts.<State3, input3>

<State2, input2>

<State1, input1>

<State3, input3>

<State2, input2>

<State1, input1>

<State3, input3>

<State2, input2>

<State1, input1>

<State2, input2><State1, input1>

Implementation• Exactly 1 instance• Self-stabilization• Mobility

• Where and when to move? – Can be decided by current state

– Ensure the right order of events

– Ensure nodes´ proper join/leave

x1 x2 x3

move to x2 on t2move to x3 on t3<join, id 37634>

<State, input>

• Esure that:

• Old nodes remain participants

• Enough nodes near the new

location can receive notification

• And, mobile nodes can join

Discussion

• Described how to implement a single AVMN– Can implement multiple AVMNs using the same techniques.

• There are a number of ways to optimize – Use min amount of power to reach everyone. – Use nodes that are closer to the new AVMN centrum. – If possible, take advantage of nodes movement.

Thank you

Your attention is appreciated