TexPoint fonts used in EMF.

Post on 22-Feb-2016

30 views 0 download

Tags:

description

Deterministic Multi-Channel Information Exchange. TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box .: A A A A A A A A A A. n:= # nodes. Problem :. n:= # nodes k:= # information. Problem :. Have information. ?. Disseminate to all!. - PowerPoint PPT Presentation

transcript

ETH Zurich – Distributed Computing Group Stephan HolzerETH Zurich – Distributed Computing – www.disco.ethz.ch

Stephan Holzer - ETH ZürichThomas Locher - ABB Switzerland

Yvonne Anne Pignolet - ABB SwitzerlandRoger Wattenhofer - ETH Zürich

Deterministic Multi-Channel

Information Exchange

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Problem:n:= # nodes

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Problem:n:= # nodesk:= # information Have information

Disseminate to all!?

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Problem:

Disseminate to all!?

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Problem:

Disseminate to all!?

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Problem:n:= # nodes

1

23

4

5

n

Unique IDs 1…n

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Problem:

Disseminate to all!?Easy: O(n)

Faster?

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information ExchangeI can:

send / receive

reach each node

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information ExchangeI can:

send / receive

?reach each node

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

no collision detection

I can:

send / receive

reach each node

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

switch channels

no collision detection

I can:

send / receive

reach each node

101 Mhz117 Mhz132 Mhz …

synchronus

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

switch channels

no collision detection

I can:

send / receive

reach each node

complexitycomputation: freeradio: time 1

synchronus

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Time Channels[GW85]: Ω(k + log n) 1

k

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

kTime Channels

[GW85]: Ω(k + log n) 1[HPSW11]: O(k) n

k

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

kTime Channels

[GW85]: Ω(k + log n) 1[HPSW11]: O(k) n

Optimal

k

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

kTime Channels

[GW85]: Ω(k + log n) 1[HPSW11]: O(k) n

Optimal

????

k

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Range of k [1, ] ( , log n ) [log n , n]Upper boundOn channels O() O() 1

[HPSW11] - Channels needed for time O(k):

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Range of k [1, ] ( , log n ) [log n , n]Upper boundOn channels O() O() 1

[HPSW11] - Channels needed for time O(k):

Range of k [1, log n] (log n , log n loglog n)

[log n loglog n , n- log n)

[n – log n, n]

Upper boundOn channels O() O() O() 1

This paper:

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Range of k [1, ] ( , log n ) [log n , n]Upper boundOn channels O() O() 1

[HPSW11] - Channels needed for time O(k):

Range of k [1, log n] (log n , log n loglog n)

[log n loglog n , n- log n)

[n – log n, n]

Upper boundOn channels O() O() O() 1

This paper:

randomized

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Range of k [1, ] ( , log n ) [log n , n]Upper boundOn channels O() O() 1

[HPSW11] - Channels needed for time O(k):

Range of k [1, log n] (log n , log n loglog n)

[log n loglog n , n- log n)

[n – log n, n]

Upper boundOn channels O() O() O() 1

This paper:

randomized

deterministic

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Range of k [1, ] ( , log n ) [log n , n]Upper boundOn channels O() O() 1

[HPSW11] - Channels needed for time O(k):

Range of k [1, log n] (log n , log n loglog n)

[log n loglog n , n- log n)

[n – log n, n]

Upper boundOn channels O() O() O() 1

This paper:

randomized

deterministic

Optimal?

Optimal?

Optimal?

Optimal?

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Range of k [1, ] ( , log n ) [log n , n]Upper boundOn channels O() O() 1

[HPSW11] - Channels needed for time O(k):

Range of k [1, log n] (log n , log n loglog n)

[log n loglog n , n- log n)

[n – log n, n]

Upper boundOn channels O() O() O() 1Lower boundOn channels Ω() Ω() Ω(log ) 1

This paper:

randomized

deterministic

k

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Main ingredient:

Specially taylored graphs.

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Main ingredient:

Specially taylored graphs.(Inspired by use of lossless expanders in [CK08])

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Main ingredient:

Specially taylored graphs.(Inspired by use of lossless expanders in [CK08])

Topology: Still single hop.Graphs used to select channel.

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

7

6

5

4

3

2

1V

W

Bipartite :

node IDs new names

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

7

6

5

4

3

2

1V

W

Bipartite :

node IDs new names

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Matching Graphs:• Nodes in V have degee V

W

Δ

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Matching Graphs:• Nodes in V have degee • Fixed order of edges

VW

4123

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Matching Graphs:• Nodes in V have degee • Fixed order of edges

• For any of size at most k there is

at least nodes in X have a unique i-neighbor.

VW

4123

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Matching Graphs:• Nodes in V have degee • Fixed order of edges

• For any of size at most k there is

at least nodes in X have a unique i-neighbor.

7

6

5

4

3

2

1

VW

i

ihave uniquei-neighbor

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Matching Graphs:• Nodes in V have degee • Fixed order of edges

• For any of size at most k there is

at least nodes in X have a unique i-neighbor.

7

6

5

4

3

2

1

VW

i

ihave uniquei-neighbor

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Matching Graphs:• Nodes in V have degee • Fixed order of edges

• For any of size at most k there is

at least nodes in X have a unique i-neighbor.

7

6

5

4

3

2

1 1

1

1

2

1

2

2

2

VW

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Matching Graphs:• Nodes in V have degee • Fixed order of edges

• For any of size at most k there is

at least nodes in X have a unique i-neighbor.

7

6

5

4

3

2

1 1

1

1

2

1

2

2

2

VW

X

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Matching Graphs:• Nodes in V have degee • Fixed order of edges

• For any of size at most k there is

at least nodes in X have a unique i-neighbor.

7

6

5

4

3

2

1 1

1

1

1

VW

X

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Matching Graphs:• Nodes in V have degee • Fixed order of edges

• For any of size at most k there is

at least nodes in X have a unique i-neighbor.

7

6

5

4

3

2

1 1

1

1

1

VW

XBAD

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Matching Graphs:• Nodes in V have degee • Fixed order of edges

• For any of size at most k there is

at least nodes in X have a unique i-neighbor.

7

6

5

4

3

2

1 1

1

1

2

1

2

2

2

VW

X

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Matching Graphs:• Nodes in V have degee • Fixed order of edges

• For any of size at most k there is

at least nodes in X have a unique i-neighbor.

7

6

5

4

3

2

122

2

2

VW

X

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Matching Graphs:• Nodes in V have degee • Fixed order of edges

• For any of size at most k there is

at least nodes in X have a unique i-neighbor.

7

6

5

4

3

2

122

2

2

VW

XGOOD

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Matching Graphs:• Nodes in V have degee • Fixed order of edges

• For any of size at most k there is

at least nodes in X have a unique i-neighbor.

7

6

5

4

3

2

1 1

1

1

1

VW

X

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Matching Graphs:• Nodes in V have degee • Fixed order of edges

• For any of size at most k there is

at least nodes in X have a unique i-neighbor.

7

6

5

4

3

2

1 1

1

1

1

VW

X

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Matching Graphs:• Nodes in V have degee • Fixed order of edges

• For any of size at most K there is

at least nodes in X have a unique i-neighbor.

exist if 7

6

5

4

3

2

1 1

1

1

2

1

2

2

2

VW

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

What are these graphs good for?

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

What are these graphs good for?

Renaming

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

What are these graphs good for?

Renaming

7

6

5

4

3

2

1 1

1

1

2

1

2

2

2

VW

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

What are these graphs good for?

Renaming

• To each of the k «reporters» we can assing a new unique name in |W| in time O( using |W| channels.

7

6

5

4

3

2

1 1

1

1

2

1

2

2

2

VW

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

What is renaming good for?

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

What is renaming good for?Assignment of reporters to channels!

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

What is renaming good for?Assignment of reporters to channels!

Example: k < log n

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

What is renaming good for?Assignment of reporters to channels!

Example: k < log n

n

Original names

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

What is renaming good for?Assignment of reporters to channels!

Example: k < log n

n

Original names

O()

New namesof reporters

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n O()

Original names

New namesof reporters

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

n O()

Original names

New namesof reporters

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Time: O( k )

𝑆𝑖𝑧𝑒 :𝑛 /2n O()

Original names

New namesof reporters

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

𝑆𝑖𝑧𝑒 :𝑛 /2n O()

Original names

New namesof reporters

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

𝑆𝑖𝑧𝑒 :𝑛 /2n O()

Original names

New namesof reporters

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

𝑆𝑖𝑧𝑒 :𝑛 /2

Send on channel “new name” .

n O()

Original names

New namesof reporters

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

𝑆𝑖𝑧𝑒 :𝑛 /2

Send on channel “new name” .Send on channel “new name” .

n O()

Original names

New namesof reporters

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Send on channel “new name” .Send on channel “new name” .

𝑆𝑖𝑧𝑒 :𝑛 /2

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Send on channel “new name” .Send on channel “new name” .

𝑆𝑖𝑧𝑒 :𝑛 /2 Example: 3 channels

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Channel 3

Example: 3 channels

Channel 1

{1,2}{1,3 }{2,3 }

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Channel 3

Example: 3 channels

Channel 1Send 2 times

{1,2}{1,3 }{2,3 }

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Channel 3

Example: 3 channels

Channel 1Send 2 times

{1,2}{1,3 }{2,3 }

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

{1,2}{1,3 }{2,3 }

n:= # nodesk:= # information

Channel 3

Example: 3 channels

Channel 1Send 2 times

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Channel 3

Example: 3 channels

Channel 1Send 2 times

{1,3 }

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Unique Subset

Channel 3

Example: 3 channels

Channel 1Send 2 times

{1,3 }

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Unique Subset

Channel 3

Example: 3 channels

Channel 1Send 2 times

{1,3 }

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n:= # nodesk:= # information

Unique Subset

Channel 3

Example: 3 channels

Channel 1Send 2 times

{1,3 }

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

Unique Subset

Channel 3

Example: 3 channels

Channel 1Send k times

{1,3 } O( k )

n:= # nodesk:= # information

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

O( k )

Send on channel “new name” .Send on channel “new name” .

Range of k [1, log n] (log n , log n loglog n)

[log n loglog n , n- log n)

[n – log n, n]

Upper boundOn channels O() O() O() 1Lower boundOn channels Ω() Ω() Ω(log ) 1

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

n O()

Original names

New namesof reporters

O() channels

ETH Zurich – Distributed Computing Group Stephan Holzer

Deterministic Multi-Channel Information Exchange

in Summary … Detect / Disseminate Information!

101 Mhz117 Mhz132 Mhz …𝜣 (𝒌)

{1,3 }7

6

5

4

3

2

11

1

1

2

1

2

2

2

VW

X

Range of k [1, log n] (log n , log n loglog n)

[log n loglog n , n- log n)

[n – log n, n]

Upper boundOn channels O() O() O() 1Lower boundOn channels Ω() Ω() Ω(log ) 1

ETH Zurich – Distributed Computing Group Stephan HolzerETH Zurich – Distributed Computing – www.disco.ethz.ch

Stephan Holzer - ETH ZürichThomas Locher - ABB Switzerland

Yvonne Anne Pignolet - ABB SwitzerlandRoger Wattenhofer - ETH Zürich

Thank You!Questions & Comments?

ETH Zurich – Distributed Computing Group Stephan HolzerETH Zurich – Distributed Computing – www.disco.ethz.ch

Stephan Holzer - ETH ZürichThomas Locher - ABB Switzerland

Yvonne Anne Pignolet - ABB SwitzerlandRoger Wattenhofer - ETH Zürich

Thank You!Questions & Comments?