An Arc-Path Model for OSPF

Weight Setting Problem

Dr.Jeffery Kennington
Anusha Madhavan

1
Agenda
 Introduction

 Node-Arc Model

 Arc-Path Model

 Empirical Analysis and Comparison

 Conclusion
2
Introduction
 OSPF – Open Shortest Path First
 Interior Gateway Protocol
Routing Information in an autonomous system

 Link State based Algorithm

 The state of the interface or link is used to decide the path on

which the information is routed

 Multiple links with same state is possible. Demand to a

destination can be routed on multiple paths.

3
Routing using OSPF
 Routers maintain database with link state
information, weights computed using link state, IP
address etc.
 This database in each router is updated by
transmitting Link State Advertisements throughout
the autonomous system
 A shortest path tree is constructed by each router
with itself as the root node and based on weights in
the database.

4
Illustration of OSPF
a
Traffic from a to z =1200

[Weight, Flow]
Router 1
[1, 600]

[1, 600]
Router 2 OC-12, 622 Mbps
[1, 300] [1, 300]

Router 3
Router 4 Router 5
[2,300] [3,600]
[2,300]

Router 6

OC-48, 2488 Mbps

z
5
Disadvantages
 Lack of prior knowledge of point to point demands
may result in congestion as seen in link OC-12

 Updating the weights based on the link state

information

6
Node-Arc Weight Setting
Problem
 The weight-setting problem as defined by Pioro and
Medhi is as follows:
 Given: the network topology, the link capacities, and a set of
point-to-point demands
 Find: Weights and flows for each link
 Constraints: the demands must be satisfied, the capacities
cannot be violated, and at each node the total entering flow to a
given destination is split equally among all out-going links that lie
on the shortest paths to that destination.

7
Illustration of Node-Arc Model
Routing 30 units from Node 2 to Node 7
Routing 10 units from Node 4 to 7

3 15 5 15 +5+7.5 7
1
1 1

15 5+7.5 5+7.5
1 1 1

2
15 4 5+7.5 6 8
1 1
30 10

fij lij =0
i j
wij cij=50
8
Illustration of Node-Arc Model
Routing 50 units from Node 2 to 7

1 3 25 5 50 7
1 1

25 25
1 1 2, 5

2
25 4 6 8
1 1
50

fij lij =0
i j
Wij, cij cij=50
9
Node-Arc Model Formulation
 Definition of sets, parameters and variables
 N denotes the set of nodes (routers)
 L denotes the set of links (unordered pairs of nodes)
 E denotes the set of arcs
 h od denotes the demand volume to be routed from origin o to
destination d
od
 D denotes the set of (o, d ) pairs such that h > 0
 H is the sum of demand volumes (i.e.)

H  h od

( o , d )D

10
Node-Arc Model Formulation
 Set D  d : (o, d )  D 
 cij denotes the capacity of arc (i, j )  E
 The requirement for commodity d  D at node i , denoted by
g id is defined below:
 h
( s , d )D
sd
, i d , d  D

g id  hid , (i, d )  D, d  D

0 , otherwise

11
Node-Arc Model Formulation
 wij denotes the weight on arc (i, j )  E
 xijd denotes the flow on arc with destination d  D
 rid denotes the distance from node i to node d on the
shortest path to node d  D
 y id denotes the common value of the flow assigned to arcs
originating at i and contained in shortest paths from i to
dD
d
 The binary decision variable u ij = 1 if arc (i, j ) belongs to the
shortest path to node d ; else 0

12
Illustration of Node-Arc Model
Routing 30 units from Node 2 to Node 7
Routing 10 units from Node 4 to 7
15 15 +5+7.5
1 3 5 7
1 1
15 5+7.5 5+7.5
1 1 1
15 5+7.5
2 4 6 8
1 1
30 10
7
7
y27 15 u 23 1
x 15
23 r27 3
7
y47 12.5 u 45 1
x 12.5 r47 2
7
45
7
u68 0
fij lij =0
i j
wij, cij cij=50
13
Formulation of Node-Arc
Problem
 Objective:
 w
minimize
ij
 Constraints: ( i , j )E
 The first set of constraints ensures that demand at node is
satisfied and ensures the conservation of flow at each node


( k , j )E
x kjd   ik k
x d
 g
( i , k )E
d
k  N , d  D

 The arc capacity constraints are

 ij cij
x d
(i, j )  E
d D

14
Formulation of Node-Arc
Problem (contd.)
 The third set of constraints ensures that flows on shortest paths
for each (o, d ) pair are equal.
0  y id  xijd (1  u ijd ) H d  D , (i, j )  E

 The fourth set of constraints prevents flow on any arc that is not
in a shortest path for demand node d .
xijd u ijd H d  D, (i, j )  E

 The next set of constraints ensures that the lengths of shortest

paths for an (o, d ) pair are equal.
0 r jd  wij  rid (1  u ijd ) H d  D , (i, j )  E

1  u ijd r jd  wij  rid d  D , (i, j )  E

15
Formulation of Node-Arc
Problem (contd.)
 The boundary conditions are:

wij  1 (i, j )  E

xijd  0 d  D , (i, j )  E

rid 0 i  N , d  D
y id 0 i  N , d  D

u ijd  {0,1} d  D , (i, j )  E

16
Arc-Path Weight Setting
Problem
 The objective is to split the flow equally on limited paths
of an (o, d ) demand pair
 By limiting the candidate paths, all possible flow
combinations can be enumerated
 A unique pattern number is assigned to a possible flow
distribution in the paths
 The selection of a pattern suggests if there is a single or
multiple shortest paths for a (o, d ) pair
 The weights on the arcs can be computed based on the
pattern selection subject to capacity constraints

17
Illustration of Arc-Path Model
Routing 30 units from Node 2 to Node 7
Routing 10 units from Node 4 to 7

1 3 5 7

2 4 6 8

18
 Flow Distribution Pattern for 2
(o,d) Pairs
Candidate Paths Demand (2,7) Demand (4,7)
p1 - 2->3->5->7 Path p1 p2 p3 p4 p5
Pattern
p2 - 2->4->5->7 v1 30 0 0
v2 0 30 0
p3 - 2->4->6->7 v3 0 0 30
v4 15 15 0
p 4 - 4>5->7 v5 0 15 15
v6 15 0 15
p5 - 4->6->7 v7 10 10 10
v8 10 0
Patterns v7 and v10 are v9 0 10
selected v 10 5 5

19
Illustration of Arc-Path Model
Routing 30 units from Node 2 to Node 7
Candidate Paths – 2->3->5->7; 2->4->5->7;2->4->6->7
Routing 10 units from Node 4 to 7
Candidate Paths – 4->5->7; 4->6->7
3 10 5 10 +10 +5 7
1
1 1

10 10+5 10+5
1 1 1

2
20 4 10 +5 6 8
1 1
30 10

fij lij =0
i j
wij cij=50
20
Arc-Path Model Formulation
 Definition of sets, parameters and variables
 Pod denotes the set of paths for demand pair (o, d )  D
 P = ( o,
d )D
Pod denotes the denote the set of paths computed
using the least hops criterion.
 J p denotes the arcs in path p P
 Qij denotes the paths that include arc
(i, j )  E
 Rod denotes the pattern numbers for each (o, d ) pair
 Assumptions
 The demand values for all (o, d )  D are equal
 The number of candidate paths for each demand pair is 3 (i.e.)
Pod 3 (o, d )  D
21
Arc-Path Model Formulation
(contd.)
 Definition of sets, parameters and variables
1
 Set C od denote the set of patterns that are associated with a
single path for demand pair (o, d )  D
2
 Set C od denote the set of patterns that are associated with two
paths for demand pair (o, d )  D
3
 Set C od denote the set of patterns that are associated with three
paths for demand pair (o, d )  D
Let p be the path associated with pattern v  C od and let set
1

1
Todv  p
Let pa and p b be the paths associated with each v  C od and let
2

2
Todv  p a , pb 

22
 Arc-Path Model Formulation
(contd.)
 Let pa , pb and pc be the paths associated with each v  C od
3
and
let Todv  p a , pb , p c 
3
Demand (2,7) Demand (4,7)
Path p1 p2 p3 p4 p5

C271 v1 v2 v3  T271 v1  p1, T271 v2  p2 , T271 v3  p3  Pattern

v1 30 0 0
C v8 v9  T
1
47
1
47 v8  p4 , T1
47 v9  p5  v2 0 30 0

C272 v4 v5 v6  T272 v4  p1 , p2 , T272 v5  p2 , p3 , T272 v6  p1 , p3  v3 0 0 30

v4 15 15 0
C v10 
2
47 T 2
47 v10  p4 , p5  v5 0 15 15

C273 v7  T273 v7  p1 , p2 , p3  v6 15 0 15

v7 10 10 10
v8 10 0
v9 0 10
v 10 5 5

23
Arc-Path Model Formulation
(contd.)
  
Let V  1, 2,  ,7 D be the set of patterns
 The flow in path p  P , pattern is stored in matrix bvp
 
Let V p  v : v  V , bvp  0
 Let f p be the flow of path p  P
 Let l p be the length of path p  P
 The binary variable qv is 1 if pattern v  V is selected; and 0,
otherwise

24
Formulation of Arc-Path
Problem
 Objective:

wij
minimize
( i , j )E
 Constraints:
 The first set of constraints ensures that that only one
pattern is selected for each (o, d ) pair.

q
vRod
v  1 (o, d )  D

 The arc capacity constraints are

 b
pQij vV p
vp qv  cij (i, j )  E

25
Formulation of Arc-Path
Problem (Contd.)
 The third set of constraints calculates the length of each path

w
( i , j )J p
ij l p p  P

 The fourth set of constraints guarantee that the weights on arcs

(i, j ) and ( j , i ) are equal
wij  w ji (i, j )  L

 The following sets of constraints ensure that if a pattern with flow

on a single path is chosen then the length of that path is shortest.
1 1 1
l p  (lt  1)  M (1  qv ) (o, d )  D, v  Cod , p  Todv , t  Pod \ Todv

26
Formulation of Arc-Path
Problem (Contd.)
 The following sets of constraints ensure that if a pattern with flow
on two paths is chosen then the lengths of those two paths are
shortest
l p  (lt  1)  M (1  qv ) (o, d )  D, v  Cod2 , p  Todv
2 2
, t  Pod \ Todv

 The last sets of constraints ensure that the lengths of multiple

paths with flow are equal
2 2 2
l p  lt  M (1  qv ) (o, d )  D, v  Codv , p  Todv , t  Todv , p t

3 3 3
l p  lt  M (1  qv ) (o, d )  D, v  Codv , p  Todv , t  Todv , p t

27
Formulation of Arc-Path
Problem (Contd.)
 The boundary conditions are given below:

0  wij  216 and integer  (i, j )  E

f p 0, l p 0 p  P
qv  {0,1} v  V

28
Empirical Analysis
 Comparison between node-arc and arc-path models

 The test cases were generated from 6 different networks

 The demand value was fixed at 10 units

 The capacity on the arcs were generated randomly in the range

[50,100]

 Each test case had a maximum of 2 hours to compute weights

on the arcs

29
Summary and Conclusions
 OSPF algorithm involves developing a shortest path tree
by each router with itself as the root node
 Data packets to any other router are directed along this
shortest path tree.
 In this approach the point-to-point demands are
disregarded
 A node-arc based integer programming model to
determine the optimal weights for a given problem
instance was presented
 The node-arc model balances flow by splitting the
incoming flow at a node equally among the outgoing arcs.
30
Summary and Conclusions
 The node-arc model is very difficult to solve
 An improved alternative model using an arc-path
approach was presented
 The arc-path model splits the flow at origin node equally
among all the outgoing paths.
 The arc-path approach allows the user to restrict the
number of candidate paths
 Restricting the solution space allows much larger
problems to be solved using the arc-path approach

31