Virtual Infrastructure for Collision-Prone Wireless Networks

1
Virtual Infrastructure for Collision-Prone
Wireless Networks

• Seth Gilbert
• Gregory Chockler Nancy
  Lynch

2
Wireless Ad Hoc Networks
Lots of devices Lots of applications
Applications Environmental monitoring Border monitoring Messaging Fire detection Automotive networks Firefighting
3
Wireless Ad Hoc Networks
Lots of complications

• Unreliable communication
• Contention
• Collisions
• Noise
• Lost messages

• Unknown Availability
• Fault-prone devices
• Ad hoc deployments
• Unknown topology
• Unknown participants
• Mobility
• Dynamic joining/leaving

4
Fixed Infrastructure
For Example
cell towers
base stations

• Simplifies wireless networks
• Communication
• Avoids contention and collisions
• Unknown availability
• Reliable overlay
• Known topology
• Mobility
• Fixed and predictable

servers
5
Virtual Infrastructure
Unreliable ? Reliable Ad hoc ? Fixed net
6
Todays Topics

• Introduction
• Modeling
• Emulating a virtual node
• Convergent History Agreement
• Emulating virtual infrastructure

7
Wireless Networks
Overview

• Mobile nodes
• Crash failures
• Location updates
• Unreliable wireless broadcast
• Synchronous
• Collision detection
• Contention manager

8
Wireless Networks

• Broadcast within some radius R.
• Interference up to some radius R.

Local Wireless Broadcast
R
R
9
Wireless Networks

• Eventually, if only one nearby node broadcasts in
  a round, then its message will be delivered.

Eventual Collision Freedom
10
Wireless Networks
Collision Detectors

• Bob either
• Gets both messages
• or
• Detects a collision

Bob
Collin
Alice
CD

• ?A-C complete, eventually accurate
• Detect if any message is lost.
• Eventually no false positives (after
  stabilization).

11
Collision-Prone Network
Contention Manager(s)

• Eventually reduces contention/collisions
• Example backoff protocol
• Advice only nodes can ignore.
• Eventually, if only 1 node broadcasts, then no
  collisions (after stabilization).

heads ? Awake keep trying to broadcast.
tails ? Asleep wait.
silent round ? Awake start again.

• Each node contends when it wants to broadcast.
• Each contention manager outputs advice
• active it is safe for the mobile node to
  broadcast
• passive the mobile node should not broadcast.

12
Wireless Networks
Contention Manager Properties

• Eventually non-interfering
• Eventually, nearby nodes are not advised to be
  active.
• Eventually regionally fair
• Eventually, if
• Only nearby nodes contend.
• Any non-failed node contends for sufficiently
  long.
• Then
• Some non-failed, contending node is advised to be
  active for sufficiently long.

13
Wireless Networks
Overview

• Mobile nodes
• Crash failures
• Location updates
• Unreliable wireless broadcast
• Synchronous
• Collision detection
• Contention manager

14
Todays Topics

• Introduction
• Modeling
• Emulating a virtual node
• Basic Idea
• Convergent History Agreement (CHA)
• CHA Protocol
• Emulating virtual infrastructure

15
Building Virtual Infrastructure
Basic Idea Replicated State Machine
16
Building Virtual Infrastructure
Basic Idea Replicated State Machine
17
Building Virtual Infrastructure
Basic Idea Replicated State Machine

1. Leader / backup
2. Leader sends receives messages for the virtual
   node

1. Each participant is a replica.
2. Replicas execute a consistency protocol

18
Consistency First Attempt
Consensus as a building block.

• For each virtual round
• Run consensus.
• Replicas agree on
• Messages for virtual node to receive.
• Message for virtual node to send.
• State update.
• Update state.
• Leader sends message for virtual node.

19
First Attempt
Consensus as a building block.

• Problems
• Consensus does not terminate until collisions
  stop.
• Virtual rounds are non-constant sized
• Rounds are unsynchronized!
• Virtual nodes do not detect collisions
• Delay until collisions stop.

20
Goals
Mobile nodes, virtual nodes

• Can communicate with both real and virtual nodes.
• Detect collisions when a message is lost.
• Emulation of virtual nodes is transparent.
• Rounds are synchronized
• Rounds are constant length
• In each round, replicas have a consistent view
• Constant-sized messages

Virtual round emulation
21
Convergent History Agreement

• Sequence of instances
• For each instance k
• Each node proposes an input.
• Each node receives a history output h, or .

CHA k4
h1
propose(x)
h2
propose(y)
propose(z)
CHA k7
CHA k4
CHA k1
CHA k2
CHA k3
CHA k5
CHA k6
22
Convergent History Agreement

• A history is a sequence of proposals, one per
  instance, e.g.
• Validity Each history contains only real
  proposals.
• Agreement Every pair of histories shares a
  common prefix of proposals/non-proposals.
• Liveness Eventually, every instance outputs
  histories, and every element in the history is a
  proposal.

CHA k7
CHA k4
CHA k1
CHA k2
CHA k3
CHA k5
CHA k6
23
Emulating a Virtual Round
Using Convergent History Agreement

• Determine proposals.
• Mobile nodes broadcast their messages for the
  virtual round.
• Calculate virtual nodes broadcast message, using
  most recent history, extended by collisions.
• Execute convergent history agreement (CHA).
• If CHA outputs a history, then
• Simulate receiving virtual nodes broadcast.
• Otherwise
• Report a collisions.

24
CHA Protocol
Three phases

• Each client begins with a proposal for the
  current instance.
• Round 1 Ballot
• If adviceactive then broadcast(ballot)
• Round 2 Veto-1
• If collision then broadcast(veto)
• Round 3 Veto-2
• If collision then broadcast(veto)
• Update state

25
Update State
Accept
Reject
26
Update State
Accept
Tentative Accept
Tentative Reject
Reject
27
Update State
Accept
Tentative Accept
Tentative Reject
Reject

• ballotproposal, last green/yellow round

28
Ballot history

• Reconstructing the history
• Example

Ballot 9 C, 7
Ballot 8
Ballot 7 D, 4
Ballot 6 D, 4
Ballot 5 A, 3
Ballot 4 O, 1
Ballot 3
Ballot 2 R, 1
Ballot 1 P, 0
29
Ballot history

• Reconstructing the history
• Example

Ballot 9 C, 7
Ballot 8
Ballot 7 D, 4
Ballot 6 D, 4
Ballot 5 A, 3
Ballot 4 O, 1
Ballot 3
Ballot 2 R, 1
Ballot 1 P, 0
30
Ballot history

• Reconstructing the history
• Example

Ballot 9 C, 7
Ballot 8
Ballot 7 D, 4
Ballot 6 D, 4
Ballot 5 A, 3
Ballot 4 O, 1
Ballot 3
Ballot 2 R, 1
Ballot 1 P, 0
31
Ballot history

• Reconstructing the history
• Example

Ballot 9 C, 7
Ballot 8
Ballot 7 D, 4
Ballot 6 D, 4
Ballot 5 A, 3
Ballot 4 O, 1
Ballot 3
Ballot 2 R, 1
Ballot 1 P, 0
32
Ballot history

• Reconstructing the history
• Example

Ballot 9 C, 7
Ballot 8
Ballot 7 D, 4
Ballot 6 D, 4
Ballot 5 A, 3
Ballot 4 O, 1
Ballot 3
Ballot 2 R, 1
Ballot 1 P, 0

• History
• 1. propose(P)
• 2. propose(O)
• 3. propose(D)
• 4. propose(C)
• gt PODC

33
CHAP Output

• If round is green
• Output history.
• Include proposal where indicated, otherwise.
• Else if round is not green
• Output .

34
Analysis Key Idea (1)
The color at any two replicas differs by at most
one shade.

• If some instance is green, then all accept that
  instances ballot.
• All histories agree on the same execution.
• If a history is output for some instance, then
  all replicas receive the proposal for that
  instance.
• If a ballot is red, then all the replicas reject
  it.
• If a proposal is not received, then no history
  outputs it.

35
Analysis Key Idea (2)
After the network, collision detector, and
contention manager stabilize Eventually, every
round is green.

• When the underlying network is well-behaved, then
  there are no vetos.
• Used to prove that the virtual node satisfies
• Eventual collision freedom.
• Eventual accuracy of the collision detector.
• Contention manager eventually well-behaved.

36
Todays Topics

• Introduction
• Modeling
• Emulating a virtual node
• Convergent History Agreement
• Emulating virtual infrastructure

37
Virtual Infrastructure
More than one virtual node

• Problems
• Virtual nodes communicate with each other.
• Sender adds message to send to proposal.
• Receiver adds messages to receive to proposal.
• Both run agreement.
• If sender-agreement fails, then receiver cannot
  receive the message.
• Too late!!
• Virtual node emulators may interfere with each
  other.
• Contention managers are insufficient.

38
Virtual Infrastructure
Scheduling virtual nodes

• Choose a schedule for all the virtual nodes
• Non-conflicting no neighboring virtual nodes
  scheduled at the same time.
• Fair every virtual node is scheduled.
• Use schedule to avoid contention in the
  emulation.
• Not sufficient to emulate each virtual node
  according to the schedule, since virtual nodes
  communicate.

39
Virtual Infrastructure

• Four part protocol
• Data phases
• Client phase clients broadcast messages to vns,
  each other.
• VN phase representative emulator sends msg for
  vn.
• Scheduled Agreement Instance
• Scheduled virtual nodes run the 3 phase agreement
  protocol, all at once.
• Unscheduled Agreement Instance
• Unscheduled virtual nodes run the 3 phase
  agreement protocol.
• Lasts for s 2 rounds, where s is the size of the
  schedule.
• Join Reset phases.

40
Virtual Infrastructure
Emulation results

• Bounded round emulation
• 10 (size of schedule)
• Constant overhead
• After stabilization, constant message overhead.
• Can be optimized further.

41
Ongoing and Future Work
Interesting questions
42
Ongoing and Future Work
Practical issues

• Optimization
• Smaller messages
• Less redundant communication
• Shorter virtual rounds
• Implementation aspects
• Collision detectors
• Contention managers
• Synchronization

43
Ongoing and Future Work
Applications

• Coordination Problems
• Traffic Coordination
• Air traffic control
• Rescue worker / military scenarios
• Control Problems
• Actuated sensors
• Real-time applications
• Mobile Sensors
• Floating / submergible devices
• Zebranet-like scenarios

44
Ongoing and Future Work
Extensions

• Malicious Devices
• Can we emulate virtual infrastructure in the
  presence of non-cooperative mobile nodes?
• Can we keep the state of the system secret from
  the participants?
• Cryptography
• Frequency hopping

45
Ongoing and Future Work
Extensions

• Energy Efficiency
• How efficiently can we implement virtual
  infrastructure?
• Mobile nodes share the work of implementing the
  virtual nodes.
• More efficient replication via coding theory
  techniques?

46
Summary

• Virtual Infrastructure
• Simple abstraction for
  dependable ad hoc networks
• Convergent History Agreement
• CHA Protocol / VI Emulation
• Collision detectors, contention managers
• Efficient constant overhead, constant time

47
The End
48
The End
49
(No Transcript)