49
1 Routing without Flow Control Costas Busch Rensselaer Polytechnic Institute Maurice Herlihy Brown University Roger Wattenhofer Microsoft Research

The network: mesh

  • Upload
    gaura

  • View
    43

  • Download
    0

Embed Size (px)

DESCRIPTION

Routing without Flow Control Costas Busch Rensselaer Polytechnic Institute Maurice Herlihy Brown University Roger Wattenhofer Microsoft Research. The network: mesh. Discrete time Bi-directional links At most one packet per link direction. Dynamic Routing:. - PowerPoint PPT Presentation

Citation preview

Page 1: The network:            mesh

1

Routing without Flow Control

Costas BuschRensselaer Polytechnic Institute

Maurice HerlihyBrown University

Roger WattenhoferMicrosoft Research

Page 2: The network:            mesh

2

n

n

The network: meshnn

• Discrete time• Bi-directional links• At most one packet per link direction

Page 3: The network:            mesh

3

Dynamic Routing:Packets are injected continuously

destination

Page 4: The network:            mesh

4

A new packet can be injected whenthere is a free link:

A link direction is empty

Page 5: The network:            mesh

5

Most dynamic routing algorithmsuse flow control:

Don’t utilize all the free links

Disadvantage: Network is under-utilized

Page 6: The network:            mesh

6

Our Routing Algorithm:

•No flow control

•Utilizes all the free links

Advantage: Network is fully-utilized

Page 7: The network:            mesh

7

Features of our algorithm:• Dynamic

• Hot potato

• Optimal delivery time:

• Injection time guaranty: )(nO

)(nO

Page 8: The network:            mesh

8

Talk Outline The Algorithm

Time AnalysisStabilityFuture Work

Page 9: The network:            mesh

9

Hot-Potato Routing:

• Nodes are buffer-less

• Packets are immediately forwarded

Page 10: The network:            mesh

10

Conflicts

Page 11: The network:            mesh

11

Conflict

Conflicts

Page 12: The network:            mesh

12

Deflected

Conflicts

Page 13: The network:            mesh

13

Packet states:

Running

Excited

Active

Sleeping

Priorities:

high

low

Page 14: The network:            mesh

14

Sleeping packet

destination

Random destination

Page 15: The network:            mesh

15

Sleeping packet

Follows a path to destination

destination

Page 16: The network:            mesh

16

Sleeping packet

becomes Activewithprobability

n1

n

n

Page 17: The network:            mesh

17

Active packet

Follows a greedy path

Page 18: The network:            mesh

18

Active packet

Follows a greedy path

Page 19: The network:            mesh

19

Active packet

A conflict situation

Page 20: The network:            mesh

20

Conflict

Active packet

A conflict situation

Page 21: The network:            mesh

21

Deflected

Active packet

A conflict situation

Page 22: The network:            mesh

22

Deflected

Active packet

A conflict situation

becomes Excitedwithprobability

n

p 1

Page 23: The network:            mesh

23

Excited packet

Follows a one-bend path

Page 24: The network:            mesh

24

Excited packet

Follows a one-bend path

becomes Running

Page 25: The network:            mesh

25

Running packet

Follows a one-bend path

Page 26: The network:            mesh

26

Talk Outline The Algorithm

Time AnalysisStabilityFuture Work

Page 27: The network:            mesh

27

Good condition for a column: at most non-sleeping packets with destination in the column

n10

Page 28: The network:            mesh

28

Expected delivery time for one packet: n

(when the destination column is in good condition)

Page 29: The network:            mesh

29

Initially a packet is sleeping

In expected time stepsbecomes active

n

We will show: An active packet is delivered in expected time steps nO

Page 30: The network:            mesh

30

Interrupting a one-bend path

Excited

Time 1

Page 31: The network:            mesh

31

Interrupting a one-bend path

Running

Time 2

Page 32: The network:            mesh

32

Interrupting a one-bend path

Running

Excited

Time 2

Page 33: The network:            mesh

33

Interrupting a one-bend path

Running

Running

Time 3

Page 34: The network:            mesh

34

Interrupting a one-bend path

Running

Running

Time 4

conflict

Page 35: The network:            mesh

35

Interrupting a one-bend path

Active

Running

Time 5

deflected

Page 36: The network:            mesh

36

No interruption probability:

Number of non-sleeping packetswith destinations in same column

mp)1(

Excitement probability

Page 37: The network:            mesh

37

No interruption probability:

cp m )1(

n1

n

constant

(when the destination column is in good condition)

Page 38: The network:            mesh

38

Probability of success after a deflection:

n

cp 1

Expected number of deflections until success: n

Expected delivery timefor an active packet: nO

Page 39: The network:            mesh

39

Talk Outline The Algorithm

Time AnalysisStabilityFuture Work

Page 40: The network:            mesh

40

Divide time in time periods:

n6t

Examine the condition of a column

Page 41: The network:            mesh

41

Good condition Bad condition

ne1

ne

nm 10 nm 10

1 time period

Page 42: The network:            mesh

42

Good condition Bad condition

ne1

ne

nm 10 nne1

nne

nm 10

4n time periods

1 time period

Page 43: The network:            mesh

43

Good condition Bad condition

ne1

ne

nm 10 nm 10

1 time period

Page 44: The network:            mesh

44

Proof OutlineIn a time period:

•At most new non-sleeping packets are generated with destinations in the column

n2

•At least non-sleeping packets are delivered (if )

n2nm 8

Page 45: The network:            mesh

45

Good condition Bad conditionnm 10 nne1

nne

nm 10

4n time periods

Page 46: The network:            mesh

46

Proof OutlineIn a time period:

•At most new non-sleeping packets are generated with destinations in the column

n2

•At least non-sleeping packets are delivered

n3

Page 47: The network:            mesh

47

•Most of the time, the columns are in good condition

•Each packet is delivered in expected time

Consequences:

n

Page 48: The network:            mesh

48

Talk Outline The Algorithm

Time AnalysisStabilityFuture Work

Page 49: The network:            mesh

49

•Arbitrary network topologies

•De-randomization:Determistic destinationsNo randomized algorithm

• Small number of packets