Outline Max Flow Algorithm Model of Computation Proposed Algorithm Self Stabilization Contribution 1 A self-stabilizing algorithm for the maximum flow.

1

Outline

Max Flow Algorithm

Model of Computation

Proposed Algorithm

Self Stabilization

Contribution

A self-stabilizing algorithm for the maximum flow

problem

Presented ByNiranjan

Sukumar Ghosh, Arobinda Gupta and Sriram V. Pemmaraju

6th May 2004

2

Outline

Max Flow Algorithm

Model of Computation

Proposed Algorithm

Self Stabilization

Contribution

Paper Outline

• Introduction to maximum flow problem

• Model of computation• Maximum flow algorithm for acyclic

graphs• Proof of correctness• Experimental Evaluation• Conclusion – Failure Model,

Algorithm for arbitrary graph

3

Outline

Max Flow Algorithm

Model of Computation

Proposed Algorithm

Self Stabilization

Contribution

Algorithms for Max Flow

• Sequential Algorithms– Ford and Fulkerson : Picks augmented paths

arbitrarily– Edmonds and Karp : Use BFS to construct augmented

paths– Goldmerg and Tarjan : Works in localized manner -

O(ne log(n2/e))

• Parallel Algorithms– Shiloach and Vishkin : O(n2logn) using O(n) processors

• Distributed Algorithms– Based mainly on the Glodmerg and Tarjan

• Self-Stabilized Algorithms– Distributed Algorithms

• Fault Tolerant• Adjust to dynamic changes in the network topology

4

Outline

Max Flow Algorithm

Model of Computation

Proposed Algorithm

Self Stabilization

Contribution

Self-Stabilization

• Introduced by Dijkstra [1974]• S is self-stabilizing with respect to

predicate P if it satisfies the following two properties:– Closure: P is closed under the execution of S.– Convergence: Starting from an arbitrary

global state, S is guaranteed to reach a global state satisfying P within a finite number of state transitions

• Failure Model : Transient Failure– An event that may the change the state of

the system, but not its behavior

5

Outline

Max Flow Algorithm

Model of Computation

Proposed Algorithm

Self Stabilization

Contribution

Preliminaries

• Skew Symmetry• Capacity Constraint• Residual Graph• Residual Capacity• Feasible flow• Max-flow, Min-cut

theorem

These terms are used in the same context

6

Outline

Max Flow Algorithm

Model of Computation

Proposed Algorithm

Self Stabilization

Contribution

Model of Computation

• Each node i in G corresponds to a process called process i that executes a program asynchronously

• Each edge (i, j) corresponds to a bidirectional link b/w process i and process j

• Process i local variables can be read by its neighbors but written only by process i

• Program Model (expressed using guarded commands [Dijkstra 1975])

7

Outline

Max Flow Algorithm

Model of Computation

Proposed Algorithm

Self Stabilization

Contribution

Maximum Flow Algorithm for Acyclic Graphs

• G is a acyclic digraph• For each edge (i,j), f(i,j): current flow

from node i to node j

• Both process i and process j can read from and write into the variable f(i,j)

• Each node i contains a single variable d(i): length of the shortest path form s to i in the residual graph Gf D(i): [0…n]

8

Outline

Max Flow Algorithm

Model of Computation

Proposed Algorithm

Self Stabilization

Contribution

Key Idea…

• For any node i ,– Demand(i) = Of(i) – If(i)

• Demand(t) = Infinity

• Each node i tries to restore the flow conservation constraint demand(i)=0 by– Reducing its inflow if demand(i) < 0– Increasing its inflow and or reducing its

outflow if demand(i) > 0

• Each node with positive demand attempts to pull flow via a shortest path from s to itself in Gf

• Use BFS to keep track of shortest paths from s to all nodes in Gf

9

Outline

Max Flow Algorithm

Model of Computation

Proposed Algorithm

Self Stabilization

Contribution

Notation

• Distinguished Nodes Node s remains idle with d(s) = 0 Node t executes the same program as other processes with demand(t) = INFINITY

i

bc

ay

x

Residual Graph

d = 5

d = 3d = 4

d = 5

IN(i) = {a, b, c}

D(i) = {4, 5, 6}

10

Outline

Max Flow Algorithm

Model of Computation

Proposed Algorithm

Self Stabilization

Contribution

Algorithm

• S1: Each node i, computes its d-value by examining the values d(j) for all (j,i)=Ef. d(i) = min{D[i], n}

• S2: If demand(i)<0, then total flow along incoming edges in G is reduced irrespective of d-values

• S3: if demand(i)>0 and d(i)<n, then i pulls flow along an incoming edges (j,i)=Ef and d(j)=d(i)-1

• S4: if demand(i) and d(i)=n, then there is no path from s to i in Gf and it reduces the outflow.

11

Outline

Max Flow Algorithm

Model of Computation

Proposed Algorithm

Self Stabilization

Contribution

Algorithm

• Assumption: f(i,j) never exceeds C(i,j) - New action (A5) that appropriately reduces the flow on an incident edge that has flow in excess of capacity

12

Outline

Max Flow Algorithm

Model of Computation

Proposed Algorithm

Self Stabilization

Contribution

Example

Node: a Guard: S2

Node: b Guard: S1

Node: t Guard: S3

Node: b Guard: S3

C(s,a) = C(b,t) = 2

C(a,b) = 1

13

Outline

Max Flow Algorithm

Model of Computation

Proposed Algorithm

Self Stabilization

Contribution

Example …

Node: a Guard: S3

Node: b Guard: S1

Node: b Guard: S4

Node: t Guard: S1

C(s,a) = C(b,t) = 2

C(a,b) = 1

14

Outline

Max Flow Algorithm

Model of Computation

Proposed Algorithm

Self Stabilization

Contribution

Results

15

Outline

Max Flow Algorithm

Model of Computation

Proposed Algorithm

Self Stabilization

Contribution

Contribution of this paper

• First distributed self-stabilizing algorithm for max-flow

• Inherently tolerant to transient faults• Automatically adjust to topology

changes– Arbitrary addition or deletion of edges– Addition and deletion of nodes provided that

#nodes in the network is bounded– Arbitrary changes in the capacities of the

edges

• Requires O(n2) in average case settings

16

Outline

Max Flow Algorithm

Model of Computation

Proposed Algorithm

Self Stabilization

Contribution

Thank You ….