Bit Dissemination Problem F O O D ! Bit Dissemination Problem - - PowerPoint PPT Presentation

bit dissemination problem
SMART_READER_LITE
LIVE PREVIEW

Bit Dissemination Problem F O O D ! Bit Dissemination Problem - - PowerPoint PPT Presentation

Self-Stabilizing Broadcast with O(1)-Bit Messages Emanuele Natale joint work with Lucas Boczkowski and Amos Korman 4th Workshop on Biological Distributed Algorithms (BDA) July 25-29, 2016 Chicago, Illinois preprint


slide-1
SLIDE 1

Self-Stabilizing Broadcast with O(1)-Bit Messages⋆

Emanuele Natale†

joint work with

Lucas Boczkowski∗ and Amos Korman∗ 4th Workshop on Biological Distributed Algorithms (BDA)

July 25-29, 2016 Chicago, Illinois

† ∗ ⋆preprint at goo.gl/ETNc64

slide-2
SLIDE 2

Self-Stabilizing Broadcast with O(1)-Bit Messages⋆

Emanuele Natale†

joint work with

Lucas Boczkowski∗ and Amos Korman∗ 4th Workshop on Biological Distributed Algorithms (BDA)

July 25-29, 2016 Chicago, Illinois

† ∗ ⋆preprint at goo.gl/ETNc64

(Bit Dissemination)

slide-3
SLIDE 3

Bit Dissemination Problem

F O O D !

slide-4
SLIDE 4

Bit Dissemination Problem

F O O D !

slide-5
SLIDE 5

Examples

Flocks of birds [Ben-Shahar et al. ’10]

slide-6
SLIDE 6

Examples

Schools of fish [Sumpter et al. ’08] Flocks of birds [Ben-Shahar et al. ’10]

slide-7
SLIDE 7

Examples

Schools of fish [Sumpter et al. ’08] Insects colonies [Franks et al. ’02] Flocks of birds [Ben-Shahar et al. ’10]

slide-8
SLIDE 8

Communication Model

  • Chaotic
  • Anonymous
  • Passive
  • Parsimonious

Animal communication:

slide-9
SLIDE 9

Communication Model

PULL(h, ℓ) model [Demers ’88]: at each round each agent can

  • bserve h other agents

chosen independently and uniformly at random, and shows ℓ bits to her

  • bservers.

? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?

  • Chaotic
  • Anonymous
  • Passive
  • Parsimonious

Animal communication:

slide-10
SLIDE 10

Communication Model

PULL(h, ℓ) model [Demers ’88]: at each round each agent can

  • bserve h other agents

chosen independently and uniformly at random, and shows ℓ bits to her

  • bservers.
  • Chaotic
  • Anonymous
01001
  • Passive
  • Parsimonious

01001

Animal communication:

  • Passive
  • Parsimonious
slide-11
SLIDE 11

Communication Model

PULL(h, ℓ) model [Demers ’88]: at each round each agent can

  • bserve h other agents

chosen independently and uniformly at random, and shows ℓ bits to her

  • bservers.
  • Chaotic
  • Anonymous
01001
  • Passive
  • Parsimonious

01001

Animal communication:

  • Passive
  • Parsimonious
slide-12
SLIDE 12

Communication Model

PULL(h, ℓ) model [Demers ’88]: at each round each agent can

  • bserve h other agents

chosen independently and uniformly at random, and shows ℓ bits to her

  • bservers.
  • Chaotic
  • Anonymous
01001
  • Passive
  • Parsimonious

01001

Animal communication:

  • Passive
  • Parsimonious
slide-13
SLIDE 13

(Probabilistic) Self-Stabilization

Sources’ bits (and other agents’ states) may change in response to external environment

slide-14
SLIDE 14

(Probabilistic) Self-Stabilization

Sources’ bits (and other agents’ states) may change in response to external environment

slide-15
SLIDE 15

(Probabilistic) Self-Stabilization

Sources’ bits (and other agents’ states) may change in response to external environment

slide-16
SLIDE 16

(Probabilistic) Self-Stabilization

blue vs red: 39/14 ≈ 2.8 Sources’ bits (and other agents’ states) may change in response to external environment

slide-17
SLIDE 17

(Probabilistic) Self-Stabilization

(Probabilistic) Self-stabilizing algorithm: guarantees convergence and closure w.r.t. S (w.h.p.) configuration

  • f system

S convergence closure (Probabilistic) self-stabilization: S := {“correct configurations of the system” } (= consensus on source’s bit)

  • Convergence. From any initial configuration, the

system reaches S (w.h.p.)

  • Closure. If in S, the system stays in S (w.h.p.)
slide-18
SLIDE 18

(Self-Stab.) Bit Dissemination vs Synchronization

slide-19
SLIDE 19

(Self-Stab.) Bit Dissemination vs Synchronization

slide-20
SLIDE 20

(Self-Stab.) Bit Dissemination vs Synchronization

slide-21
SLIDE 21

(Self-Stab.) Bit Dissemination vs Synchronization

slide-22
SLIDE 22

(Self-Stab.) Bit Dissemination vs Synchronization

slide-23
SLIDE 23

(Self-Stab.) Bit Dissemination vs Synchronization

slide-24
SLIDE 24

(Self-Stab.) Bit Dissemination vs Synchronization

slide-25
SLIDE 25

(Self-Stab.) Bit Dissemination vs Synchronization

Self-stablizing algorithms converge from any initial configuration

slide-26
SLIDE 26

Self-Stabilizing Clock Sync. in the PULL Model

2-Majority dynamics [Doerr et al. ’11]. Converge to consensus in O(log n) rounds with high probability.

= ⇒

slide-27
SLIDE 27

Self-Stabilizing Clock Sync. in the PULL Model

2-Majority dynamics [Doerr et al. ’11]. Converge to consensus in O(log n) rounds with high probability.

= ⇒ = ⇒

1 1 1 1 1 1 1 1 1 1 1 1 1

slide-28
SLIDE 28

Self-Stabilizing Clock Sync. in the PULL Model

2-Majority dynamics [Doerr et al. ’11]. Converge to consensus in O(log n) rounds with high probability.

= ⇒ = ⇒

1 1 1 1 1 1 1 1 1 1 1 1 1

slide-29
SLIDE 29

Self-Stabilizing Clock Sync. in the PULL Model

2-Majority dynamics [Doerr et al. ’11]. Converge to consensus in O(log n) rounds with high probability.

= ⇒ = ⇒

1 1 1 1 1 1 1 1 1 1 1 1 1

slide-30
SLIDE 30

Self-Stabilizing Clock Sync. in the PULL Model

Public

P Emul(P) 1 1 1 1 1 1 0 1 0 1 1 1 0 1 0

Public Private

ℓ bits log ℓ + 1 bits

Message Reduction Lemma

slide-31
SLIDE 31

Self-Stabilizing Clock Sync. in the PULL Model

Public

P Emul(P) 1 1 1 1 1 1 0 1 0 1 1 1 0 1 0

Public Private

ℓ bits log ℓ + 1 bits

Message Reduction Lemma Public Private Private Private

· · ·

1 1 0 0 1 0 1 0 1 0 0 1 0 1 Syn-Clock

slide-32
SLIDE 32

Results

Theorem (Self-Stabilizing Bit Dissemination). There is a self-stabilizing Bit Dissemination protocol which converges in ˜ O(log n) rounds w.h.p using 3-bit messages. Theorem (Clock Syncronization). Syn-Clock is a self-stabilizing clock synchronization protocol which synchronizes a clock modulo T in ˜ O(log n log T) rounds w.h.p. using 3-bit messages.

slide-33
SLIDE 33

Self-Stab. Bit Diss. with 1 bit: a Candidate

  • Boosting: Update their
  • pinion with majority of

their bit and the 2 bits they pull. If they see only agents of color c for s rounds, they become c-sensitive.

  • c-sensitive: Turn into

c-frozen if see value c.

  • c-frozen keep value c for

f rounds before becoming boosting. BFS(f, s). Agents can boosting, 1/0-frozen or 1/0-sensitive.

slide-34
SLIDE 34

Self-Stab. Bit Diss. with 1 bit: a Candidate

  • Boosting: Update their
  • pinion with majority of

their bit and the 2 bits they pull. If they see only agents of color c for s rounds, they become c-sensitive.

  • c-sensitive: Turn into

c-frozen if see value c.

  • c-frozen keep value c for

f rounds before becoming boosting. BFS(f, s). Agents can boosting, 1/0-frozen or 1/0-sensitive.

slide-35
SLIDE 35

Self-Stab. Bit Diss. with 1 bit: a Candidate

  • Boosting: Update their
  • pinion with majority of

their bit and the 2 bits they pull. If they see only agents of color c for s rounds, they become c-sensitive.

  • c-sensitive: Turn into

c-frozen if see value c.

  • c-frozen keep value c for

f rounds before becoming boosting. BFS(f, s). Agents can boosting, 1/0-frozen or 1/0-sensitive.

slide-36
SLIDE 36

Self-Stab. Bit Diss. with 1 bit: a Candidate

  • Boosting: Update their
  • pinion with majority of

their bit and the 2 bits they pull. If they see only agents of color c for s rounds, they become c-sensitive.

  • c-sensitive: Turn into

c-frozen if see value c.

  • c-frozen keep value c for

f rounds before becoming boosting. BFS(f, s). Agents can boosting, 1/0-frozen or 1/0-sensitive.

slide-37
SLIDE 37

Self-Stab. Bit Diss. with 1 bit: a Candidate

  • Boosting: Update their
  • pinion with majority of

their bit and the 2 bits they pull. If they see only agents of color c for s rounds, they become c-sensitive.

  • c-sensitive: Turn into

c-frozen if see value c.

  • c-frozen keep value c for

f rounds before becoming boosting. BFS(f, s). Agents can boosting, 1/0-frozen or 1/0-sensitive.

slide-38
SLIDE 38