20
Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte MIT, CSAIL

Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

Embed Size (px)

Citation preview

Page 1: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast

Gregory Chockler

Murat Demirbas

Seth Gilbert

Nancy Lynch

Calvin Newport

Tina Nolte

MIT, CSAIL

Page 2: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

2

A better model for wireless broadcast

• For sensor-actuator networks dependability becomes crucial

• Developing reliable and provably correct algorithms for wireless ad hoc networks require a formal model of wireless broadcast

• What makes a good model ?

Usability: Simplicity & Power

Model should allow algorithm designers to ignore the low-level details

Model should allow complicated and interesting algorithms Realism: Fidelity to the real world phenomena

Algorithms designed and verified using the model should work as expected in the real world

Page 3: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

3

Myths of wireless broadcast models

1. Wireless broadcast is reliable

Origin: “Coordinated Attack” is unsolvable in the presence of unreliable channels [even a single message loss] FLP result

Reality check: Collisions are common place*

Even under low traffic loads, MAC layer (collision avoidance by back off) cannot avoid all concurrent transmissions

Hidden Node Effect

802.11, BMAC, SMAC, TMAC

Electromagnetic interference

*Understanding packet delivery performance in dense WSN, Zhao Govindan Taming the underlying challenges of multihop routing in SN, Woo et al. Experimental evaluation of wireless simulation assumptions, Kotz et al. Complex behavior at scale: An experimental study of … Ganesan et al.

Page 4: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

4

Myths of wireless broadcast models

2. Transmitter can detect collision

3. Collisions are uniform

Origin: Ethernet model

Reality check:

Transmission power swamps the receiver

Collisions are partial

Transmission is partially affected by transmissions in neighboring regions

Physical characteristics of radio broadcast

Page 5: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

5

Myths of wireless broadcast models

4. Collision detection is reliable

Origin: Ethernet model / Convenience

Reality check:

• Capture effects prevent reliable collision detection*

A strong transmission captures the receiver and a weaker transmission goes unnoticed

• Noisy channels prevent reliable collision detection

* Exploiting the Capture Effect for Collision Detection and Recovery,Whitehouse, Woo, Jiang, Polastre, Culler. EmNetsII, 2005.

Page 6: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

6

Contributions

• Partial Collision Model (PCM)

• Vote-veto algorithm for consensus in PCM**

Two phase algorithm that tolerates partial collisions Constant round consensus

• Solvability of consensus under varying quality of collision detection**

** Consensus and Collision Detectors in Wireless Ad Hoc Networks,Chockler, Demirbas, Gilbert, Newport, Nolte. To appear in PODC 05

Page 7: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

7

Outline of the talk

• Preliminaries

• Collision Detection (CD)

• Partial Collision Model (PCM)

• Vote-Veto Algorithm

• Ongoing Work

Page 8: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

8

Preliminaries

• Ad hoc network, number and id of nodes unknown

• Nodes can fail by crashing anytime

broadcast is an atomic operation (unlike in one-to)

• Synchronized rounds

Broadcast at most one message Receive messages Perform state transition

Page 9: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

9

Radio broadcast

• Basic properties

Integrity: each message received is previously broadcast No duplication: each message is received at most once

• Eventual Collision Freedom (ECF)

Exists a round r_ecf s.t. ForAll r>r_ecf, if at most b nodes broadcast in r, then all correct nodes receive all messages in r

MAC protocols can satisfy collision free period for unknown r_ecf, b To achieve ECF, we employ an active-passive service that obliviously

provides eventual good advice to nodes

Page 10: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

10

Collision detection

Necessary for coping with undetectable message loss

Receiver side monitoring and notification of collisions No info wrt # of lost messages or identities of senders

• Completeness: Ability to detect collisions

Majority-complete: a collision is detected if a majority of messages in a round is lost

0-complete: collision is detected if all messages in a round is lost

• Accuracy: No false positives

Always and eventually accurate CD

• Receiver side collision detection is easily implementable in mote and 802.11 platforms

Page 11: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

11

Partial Collision Model

• Integrity, No duplication, ECF, and CD

No restriction on the pattern of collisions

Usability:

Simple Powerful enough to solve consensus in constant rounds for

majority-complete CD and log rounds for 0-complete CD

Realism:

Both ECF and majority/0-complete CD are easily achievable in current wireless networks (Mote and 802.11)**

We have implementations for Mica2/TinyOS platform

*Exploiting the Capture Effect for Collision Detection and Recovery,

A new backoff alg for the IEEE802.11 distributed coordination fn., …

Page 12: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

12

Consensus problem

• Agreement: No two nodes in P decide differently

• Validity: Decision v is proposed by a node in P

• Termination: All correct nodes eventually decide

Single-hop consensus

All nodes in P are within single-hop Building block for multi-hop consensus

Applications:

Beam control scenario Virtual Traffic Lights scenario Reliable (multi-hop) reprogramming

Page 13: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

13

Vote-Veto algorithm

• Two phases: vote and veto

• The algorithm is adaptive, employs active-passive service

• Vote phase:

Every active node sends out its vote If a node hears no collision, the node updates its vote to minimum of

received votes If a node hears collision or different votes, it decides to veto

• Veto phase:

If no veto messages are received or collisions detected, then a node can decide, else nodes continue to next round

• Intuition: By having a dedicated veto phase, effects of collision is detectable

Page 14: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

14

Proof (for majority-complete CD)

Let r be the first round any node decides

Since no node vetoed in r, every node received only a single vote and no collision detection during vote phase in r-1

Since maj-◊AC detects when ≥ half the messages are lost, each node receives a majority of messages broadcasted in vote phase in r-1

Since every majority set intersects, all nodes received the same unique vote.

ECF, active-passive service, eventual accuracy leads to a vote phase where every participant converge to a vote. After the second vote phase no node vetoes, hence every node decides in at most 5 rounds after Earliest-Stabilization-Time.

Page 15: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

15

Consensus results overview

• Vote-Veto algorithm for AC, Maj-AC, ◊AC, ◊Maj-AC

• Three phase algorithm that iterates over each bit of the vote for 0-AC, 0-◊AC

ECF

AC Θ(1)

Maj-AC Θ(1)

0-AC Θ(log|V|)

◊AC Θ(1)

Maj-◊AC Θ(1)

0-◊AC Θ(log|V|)

Page 16: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

16

Implementation

• Mica2/TinyOS platforms

• Collision detection may be maj-complete

Detects a preamble of a message during reception of a message Courtesy of Whitehouse et al. case RX_STATE: {

...

if (((data_in == (0xaa)) || (data_in == (0x55)))) {signal CollisionDet.collision();}…

• Vote-Veto implementation is around 250 lines of NesC

• Experiments on MistLab MIT (motelab)

Page 17: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

17

Ongoing work

• We have implemented Virtual Traffic Lights on Mica2/TinyOS

Experiments with up to 15 nodes in single hop on the MistLab testbed(MIT/CSAIL) show success

Ongoing work on evaluation of CD

• Practical implementation of the Virtual Node architecture

Implementation of a reliable virtual node by means of maintaining a replicated state on unreliable, mobile nodes

• Efficient multi-hop consensus algorithms

• Extending to Byzantine faults

Page 18: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

18

Bonus material

• Consensus algorithm using 0-complete CD

• Consensus for no-collision-freedom model

Page 19: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

19

3 phase algorithm for 0-◊AC consensus

• Phase 1 :

Each active node proposes an estimate Every node adopts the minimum estimate it hears

• Phase 2 : one round for each bit in the estimate

If a node has 1 in the corresponding bit, it broadcasts a message If a node has 0, it listens & decides to veto if it hears a message or

collision

• Phase 3 :

Any node that decides to veto broadcasts a veto If any node performs a veto, nodes start from phase 1

Page 20: Reconciling the Theory and Practice of (Un)Reliable Wireless Broadcast Gregory Chockler Murat Demirbas Seth Gilbert Nancy Lynch Calvin Newport Tina Nolte

20

Consensus for No CF

ECF No CF

AC Θ(1) Θ(log|V|)

Maj-AC Θ(1) Θ(log|V|)

0-AC Θ(log|V|) Θ(log|V|)

◊AC Θ(1) Impossible

Maj-◊AC Θ(1) Impossible

0-◊AC Θ(log|V|) Impossible