space-optimal deterministic rendezvous

38
Space-Optimal Deterministic Rendezvous Stéphane Devismes VERIMAG UJF, Grenoble I Joint work with Fabienne Carrier, Yvan Rivierre (VERIMAG, UJF, Grenoble I), and Franck Petit (LIP6, UPMC, Paris 6)

Upload: elpida

Post on 13-Jan-2016

40 views

Category:

Documents


0 download

DESCRIPTION

Space-Optimal Deterministic Rendezvous. Stéphane Devismes VERIMAG UJF, Grenoble I. Joint work with Fabienne Carrier, Yvan Rivierre (VERIMAG, UJF, Grenoble I), and Franck Petit (LIP6, UPMC, Paris 6). System Settings. Graph G=(V,E) of n nodes and m bidirectional links - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Space-Optimal Deterministic Rendezvous

Space-Optimal Deterministic Rendezvous

Stéphane DevismesVERIMAG

UJF, Grenoble I

Joint work withFabienne Carrier, Yvan Rivierre (VERIMAG, UJF, Grenoble I), and

Franck Petit (LIP6, UPMC, Paris 6)

Page 2: Space-Optimal Deterministic Rendezvous

System Settings Graph G=(V,E)

of n nodes and m bidirectional links

Set of k mobile agents

Nodes are anonymous

Agents are autonomous and oblivious

Page 3: Space-Optimal Deterministic Rendezvous

System Settings The agents move asynchronously They cannot (explicitly) communicate

together (even being located at the same node)

They have no knowledge of each other, in particular they do not know k

They have no knowledge about G, in particular they know nothing about n, m, the diameter or maximum degree of G, etc.

Page 4: Space-Optimal Deterministic Rendezvous

Rendezvous

The agent are required to eventually meet and stop at the same node.

Initially, no agent is present in G Agents can be inserted at any time

Deterministic solutions

Page 5: Space-Optimal Deterministic Rendezvous

Related Works Two synchronous non oblivious agents

[Alpern 76] [De Marco et al., 06] [Kowalski and Pelc 04]

k asynchronous agents provided that k and n are coprime and the edge labeling has sense of direction [Barrière et al., 07]

k oblivious agents able to take a snapshot of the whole system in a ring [Klasing et al., 08]

Page 6: Space-Optimal Deterministic Rendezvous

Anonymous, oblivious agents

No a priori conditions on n and k

No knowledge

Impossibility Result[De Marco et al., 06][Barrière et al., 07]

2

2

2

2

2

2

1

1

1

1

1

1

Page 7: Space-Optimal Deterministic Rendezvous

Impossibility Result[De Marco et al., 06][Barrière et al., 07]

Anonymous, oblivious agents

No a priori conditions on n and k

No knowledge

2

2

2

2

2

2

1

1

1

1

1

1

Page 8: Space-Optimal Deterministic Rendezvous

Impossibility Result[De Marco et al., 06][Barrière et al., 07]

Anonymous, oblivious agents

No a priori conditions on n and k

No knowledge

2

2

2

2

2

2

1

1

1

1

1

1

Page 9: Space-Optimal Deterministic Rendezvous

Anonymous, oblivious agents

No a priori conditions on n and k

No knowledge

Impossibility Result[De Marco et al., 06][Barrière et al., 07]

2

2

2

2

2

2

1

1

1

1

1

1

Page 10: Space-Optimal Deterministic Rendezvous

Anonymous, oblivious agents

No a priori conditions on n and k

No knowledge Semi-anonymous,

oblivious agents, i.e., exactly one agent has the minimum label

Nodes equipped with whiteboards

Impossibility Result[De Marco et al., 06][Barrière et al., 07]

2

2

2

2

2

2

1

1

1

1

1

1

Page 11: Space-Optimal Deterministic Rendezvous

Contribution

Time and space complexity lower bounds

Space-optimal and asymptotically time optimal algorithm

Necessary conditions to deterministically solve the rendezvous problem

Page 12: Space-Optimal Deterministic Rendezvous

Gb

b1ub1

vb1

Lower Bounds

Any deterministic rendezvous algorithm must guarantee that at least one agent explore the whole graph.

Ga

a1a2

ak

b2

bk'

ua1

va1

Page 13: Space-Optimal Deterministic Rendezvous

Lower Bounds

Any deterministic rendezvous algorithm must guarantee that at least one agent explores the whole graph.

Any deterministic rendezvous algorithm terminates in Ω(m) rounds.

Page 14: Space-Optimal Deterministic Rendezvous

v

Lower Bounds

Any deterministic graph exploration made by an agent a terminates at the starting node of a.

v'

la

la

v

Page 15: Space-Optimal Deterministic Rendezvous

Lower Bounds

Any deterministic graph exploration made by an agent a terminates at the starting node of a.

Any deterministic rendezvous algorithm requires that each agent a writes its label la on the whiteboard.

lbla

✓ ✓ ✓ ✓ ✓ ✓

la la lblb

Page 16: Space-Optimal Deterministic Rendezvous

Lower Bounds

Any deterministic graph exploration made by an agent a terminates at the starting node of a.

Any deterministic rendezvous algorithm requires that each agent a writes its label la on the whiteboard.

Any deterministic rendezvous algorithm requires at least log(v+1) + log(Lmax) + 1 bits.

Page 17: Space-Optimal Deterministic Rendezvous

Algorithm 3 variables on the whiteboard of each

node v: Currentv ∈ {0,…,v-1} ∪ {⊥}, init. ⊥ Homev ∈ {F,T}, init. F Hostv: Set of labels

3 primitives for each agent a: Go(e): Sends a through the edge e From() ∈ {0,…,v-1} ∪ {⊥}: return the edge

from which a comes, ⊥ otherwise (initial state) Next() : Return the next edge label according

to From(), e.g., (From()+1 mod v) + 1

Page 18: Space-Optimal Deterministic Rendezvous

Algorithm Basic idea:

Each agent a tries to make the deterministic DFS traversal induced by the local labels of edges

Only the agent with the minimum label lmin eventually succeeds its traversal

The other agents eventually follow the traversal of lmin

Page 19: Space-Optimal Deterministic Rendezvous

Algorithm

1

2

2

2

1

1

1

1

2

32

3

Home=FHost=

Home=FHost=

Home=FHost=

Home=FHost=

Home=FHost=

Home=FHost=

1

3

Page 20: Space-Optimal Deterministic Rendezvous

Algorithm

1

2

2

2

1

1

1

1

2

3

Home=FHost=

Home=THost=X

Home=FHost=

Home=FHost=

Home=FHost=

Home=FHost=

1

3

32

Page 21: Space-Optimal Deterministic Rendezvous

Algorithm

1

2

2

2

1

1

1

1

1

2

3

3

Home=FHost=

Home=THost=X

Home=FHost=

Home=FHost=

Home=FHost=X

Home=FHost=

32

Page 22: Space-Optimal Deterministic Rendezvous

Algorithm

1

2

2

2

1

1

1

1

2

3

Home=THost=L

Home=THost=X

Home=FHost=X

Home=FHost=

Home=FHost=X

Home=FHost=

1

3

32

Page 23: Space-Optimal Deterministic Rendezvous

Algorithm

1

2

2

2

1

1

1

1

2

3

Home=THost=L

Home=THost=X

Home=FHost=X

Home=FHost=

Home=FHost=X

Home=FHost=L

1

3

32

Page 24: Space-Optimal Deterministic Rendezvous

Algorithm

1

2

2

2

1

1

1

1

2

3

Home=THost=L

Home=THost=X

Home=FHost=X

Home=FHost=

Home=FHost=L

Home=FHost=L

1

3

32

Page 25: Space-Optimal Deterministic Rendezvous

Algorithm

1

2

2

2

1

1

1

1

2

3

Home=THost=L

Home=FHost=L

Home=FHost=X

Home=FHost=X

Home=FHost=L

Home=FHost=L

1

3

32

Page 26: Space-Optimal Deterministic Rendezvous

Algorithm

1

2

2

2

1

1

1

1

2

3

Home=THost=L

Home=FHost=L

Home=FHost=X

Home=FHost=X

Home=FHost=L

Home=FHost=L

1

3

32

Page 27: Space-Optimal Deterministic Rendezvous

Algorithm

1

2

2

2

1

1

1

1

2

3

Home=THost=L

Home=FHost=L

Home=FHost=X

Home=FHost=X

Home=FHost=L

Home=FHost=L

1

3

32

Page 28: Space-Optimal Deterministic Rendezvous

Algorithm

1

2

2

2

1

1

1

1

2

3

Home=THost=L

Home=FHost=L

Home=FHost=L

Home=FHost=X

Home=FHost=L

Home=FHost=L

1

3

32

Page 29: Space-Optimal Deterministic Rendezvous

Algorithm

1

2

2

2

1

1

1

1

2

3

Home=THost=L

Home=FHost=L

Home=FHost=L

Home=FHost=L

Home=FHost=L

Home=FHost=L

1

3

32

Page 30: Space-Optimal Deterministic Rendezvous

Algorithm

1

2

2

2

1

1

1

1

2

3

Home=THost=L

Home=FHost=L

Home=FHost=L

Home=FHost=L

Home=FHost=L

Home=FHost=L

1

3

32

Page 31: Space-Optimal Deterministic Rendezvous

Algorithm

1

2

2

2

1

1

1

1

2

3

Home=THost=L

Home=FHost=L

Home=FHost=L

Home=FHost=L

Home=FHost=L

Home=FHost=L

1

3

32

Page 32: Space-Optimal Deterministic Rendezvous

Algorithm

1

2

2

2

1

1

1

1

2

3

Home=THost=L

Home=FHost=L

Home=FHost=L

Home=FHost=L

Home=FHost=L

Home=FHost=L

1

3

32

Page 33: Space-Optimal Deterministic Rendezvous

Algorithm

0

1

1

1

0

0

0

0

1

2

Home=THost=L

Home=FHost=L

Home=FHost=L

Home=FHost=L

Home=FHost=L

Home=FHost=L

0

2

21

Page 34: Space-Optimal Deterministic Rendezvous

Algorithm Performs a Rendezvous in θ(m) rounds. 2log(v+1) + log(Lmax) + 1 bits on each node.

Asymptotically optimal in time. Optimal in space.

Page 35: Space-Optimal Deterministic Rendezvous

Necessary Conditions

Labeled edges

Labels and whiteboards

Unique minimum label

[Barriere et al., 07]

Lemma 3

(Local) Determinism

Page 36: Space-Optimal Deterministic Rendezvous

Conclusion

Time and space complexity lower bounds

Asymptotically space and time optimal algorithm

Future Work : directed graphs

Page 37: Space-Optimal Deterministic Rendezvous

Thank you.

Page 38: Space-Optimal Deterministic Rendezvous