Traffic engineering for MPLSbased virtual private networks

1
Traffic engineering for MPLS-based virtual
private networks

• Author Chun Tung Chou
• Source COMPUTER NETWORKS
• Speaker Yong-Yuan Cheng (???)

2
Outline

• Introduction
• A multiobjective VPN traffic engineering problem
• An heuristic solution
• Simulation
• Conclusion

3
Introduction

• MPLS-based traffic engineering has been proposed
  as a solution to overcome the congestion problem
  of IP networks.
• The key idea is to use the path pinning
  capability of MPLS to direct the traffic flows
• avoid network congestion
• hot spots

4
Introduction

• There are numerous design choices in designing an
  MPLS-based traffic engineering scheme
• Centralised
• Distributed

5
Motivation

• Two common optimisation criteria have been
  proposed in the literature.
• minimise a linear function of the link bandwidth
  usage
• minimise a linear function of the link bandwidth
  usage
• may result in an uneven distribution of traffic
• minimise the maximum link utilisation
• maximise the room for traffic growth
• The network resource usage is not minimised

6
Goal

• we propose a multiobjective formulation of the
  traffic engineering problem
• takes into account resource usage
• link utilisation
• the number of LSPs.
• we propose an efficient heuristic to solve this
  problem.
• find a suitable path for each demand of each VPN

7
assume

• This computation is performed in an off-line
  centralised manner
• the NSP owns a physical network for providing the
  VPN service
• only one service class is offered by this NSP
• each VPN customer provides the NSP with a traffic
  demand matrix
• elements are the bandwidth requirement
• between an ingressegress pair of the VPN.

8
A multiobjective VPN traffic engineering problem

• we will make the assumption that the physical
  network G
• Sufficient capacity
• Meet the demands from all VPNs
• traffic engineering problem
• one of choosing a suitable physical route for
  each VPN demand
• a certain criterion is optimised

9
optimisation problems to be formulated

• allocated to any physical link does not exceed
  its physical capacity.
• we will need to ensure that the total capacity
• keep track of whether a particular potential path
  uses a certain physical link.

10
A multiobjective VPN traffic engineering problem

• Define the decision variables

• Capacity allocated on link euv ? e for the VPN
  requests

11
A multiobjective VPN traffic engineering problem

• In order to control the number of LSPs to be used
• introduce an additional set of decision

• The total number of LSPs or routes R that will be
  used to implement these VPNs

12
A multiobjective VPN traffic engineering problem

• The multiobjective programming VPN traffic
  engineering problem consists of two steps
• minimise the maximum link utilisation
• minimise the cost

13
Optimisation problem OPT1a

• min µ
• Subject to the constraints

(6)
(7)
(8)
(10)
(9)
14
Optimisation problem OPT1b
(11)
Subject to the constraints
(12)
(13)
(14)
(16)
(15)
15
Example
3
8
0.1
5
0.5
0.6
0.5
0.8
3
2
4
0.5
0.6
0.6
0
0.1
6
0.5
0.6
0.5
1
0.5
8
0.8
7
0.6
16
A continuous approximation

• The aim of this section is to formulate two
  linear programming (LP) problems
• give us an approximation of the simplified
  version
• OPT1a
• OPT1b
• Let be the aggregate demand from all VPNs for
  ingressegress pair (I , j) for service class s

treated as a constant
(17)
17
A continuous approximation

• We now define a set of continuous decision
  variables in the range 0,1

• capacity being used on physical link euv can be
  written as

(18)
18
OPT2a
min µ
(19)
Subject to the constraints
(20)
(21)
(22)
19
OPT2b
(23)
Subject to the constraints
(24)
(25)
(26)
20
Recovering the integer solution

• Let t1,,tD be a set of non-zero demands to be
  distributed over B different LSPs.
• B2
• h1,,B

21
Recovering the integer solution

• Note that B and ?hs are given by the solution of
  the optimisation problem OPT2b.

22
Recovering the integer solution

• In order to formulate this problem of sorting
  demands into the LSPs, we define binary decision
  variables

23
OPT3
(29)
subject to the constraint
(30)
(31)
24
Algorithm Greedy subset sum
25
Algorithm Heuristic for OPT3
26
Algorithm Heuristic for OPT3
27
Controlling the number of LSPs
OPT2b
Subject to the constraints
(34)
allows us to indirectly control the number
of LSPs being used.
(35)
(36)
28
Simulation

• We demonstrate the effectiveness of our
  algorithms using a network
• 17 nodes
• 58 links

29
Quality of the heuristic

• In each simulation
• M VPNs are generated
• VPN demand is chosen randomly from the range
  tmin,tmax
• The set of potential paths is all paths within a
  hop limit hmax

30
Quality of the heuristic
31
Quality of the heuristic

• If the set of potential paths is restricted to 6
  hops or less
• there are about 11,000 paths
• For the case of unrestricted hop limit
• potential paths is almost 50,000
• If we are to solve the integer programming
  problem for this case
• have over 5 million binary decision variables
• 100VPNs

32
Comparing different optimisation objectives

• there are 100 VPNs and the demand for these VPNs
  are randomly generated
• There are altogether 3 service classes
• Service Class 1
• has 6 hops or less
• Service Class 2
• have 9 hops or less
• Service Class 3
• no restriction

33
Comparing different optimisation objectives

• We will compare the effect of the choice of
  optimisation criterion on the traffic
  distribution.
• minimising the total network resource usage alone
• minimising only the maximum link utilisation
• multiobjective programming formulation

34
Comparing different optimisation objectives
35
Comparing different optimisation objectives
36
Comparing different optimisation objectives
37
Conclusion

• We demonstrate that this multiobjective
  formulation overcomes the problems of single
  objective formulations
• We have proposed an heuristic solution, which
  allows tractable solution.