0% found this document useful (0 votes)
72 views55 pages

Traffic Engineering in IPMPLS Networks

Uploaded by

Shariful Islam
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
72 views55 pages

Traffic Engineering in IPMPLS Networks

Uploaded by

Shariful Islam
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 55

Traffic engineering in IP/MPLS networks: the

TOTEM project
Seminar - Montefiore Institute

University of Liège (ULg), RUN


http://totem.run.montefiore.ulg.ac.be

Presentation : Simon Balon and Jean Lepropre

Traffic Engineering TEAM : Simon Balon - Olivier Delcourt


Jean Lepropre - Fabian Skivée – Guy Leduc

14 April 2005 - © RUN, ULG 1


Outline

 Introduction to traffic engineering (TE)


 The TOolbox for Traffic Engineering
Methods (TOTEM) project
 TE research performed in RUN
 Conclusion

14 April 2005 - © RUN, ULG 2


Outline

 Introduction to traffic engineering (TE)


 Notions
 Current IP architecture
 Limitations
 Definition of TE
 TE solutions

 The TOTEM project


 TE research performed in RUN
 Conclusion

14 April 2005 - © RUN, ULG 3


Notions

 We represent an intra-domain network


topology by a graph G(V,E), with V the set
of nodes and E the set of links.
A node represents a router.
 A link represents a connection between two
routers (optic fibers, etc.).
 Each link is associated with two values: a capacity
and a weight.

14 April 2005 - © RUN, ULG 4


Notions (2)

 The Internet is composed of a lot of domains


(called autonomous systems, ASes).
 Inter-domain network topology:

AS1 BGP AS2

Intra-domain
network
IGP topology

AS3

14 April 2005 - © RUN, ULG 5


Notions (3)

 Example of connection between a


computer of ULg and a computer of MIT.
Global Crossing

GÉANT Abilene

Belnet

ULg MIT

14 April 2005 - © RUN, ULG 6


Notions (4)

 The reality is somewhat more complicated…

A map of the global Internet in 2000.


Source: http://research.lumeta.com/ches/map/gallery/isp-ss.gif
14 April 2005 - © RUN, ULG 7
Notions (5)

 The traffic flowing through a network is represented by a


traffic matrix T, with tab the amount of traffic flowing from
node a to node b.
dst
node 1 node 2 ... node n
src

node 1 t11 t12 ... t1n

node 2 t21 t22 ... t2n

... ... ... ... ...

node n tn1 tn2 ... tnn

14 April 2005 - © RUN, ULG 8


Current IP architecture

 In the current Internet architecture, the routes


followed by packets are computed using a SPF
(Shortest Path First) algorithm and some link
weights assigned by the network administrator.
 Example:
A D Cost of the route CDG:
3 5 3+5=8

C G Cost of the route CEFG:


2+1+4=7
2 4
Route chosen by the SPF
B E F algorithm: CEFG
1

14 April 2005 - © RUN, ULG 9


Limitations

 By default, weights could be all set to:


1 (minimum hop routing)
 the inverse of the capacity of the links
 Bad routing scheme (traffic is not taken into
account): can lead to congestions!
A D
1 1 1
Route from A to G: ACDG
Route from B to G: BCDG
C G
If A and B send a lot of
1 traffic, congestions can
1 1
appear on CDG (while
B E F CEFG is not used).
1

14 April 2005 - © RUN, ULG 10


Limitations (2)

 We could take the traffic load into account, but


this would simply lead to oscillations.
 Routing decisions are based solely on the
destination addresses.
 Using the source address, we solve the preceding
problem but this doesn’t scale well!

A D
1 1 1
C’s forwarding table:

(src=A, dst=G)  D
C G
(src=B, dst=G)  E
1
1 1
One entry per source,
B E F this is too much!
1

14 April 2005 - © RUN, ULG 11


Definition of TE

 Traffic engineering involves adapting the routing of traffic


to the network conditions, with the joint goals of good
user performance and efficient use of network resources.
 For instance, one high level objective of a simple TE
technique could be to balance the load over all links.

A D

Load balancing:
C G
Find a way to balance the
traffic to G on CDG and
CEFG.
B E F

14 April 2005 - © RUN, ULG 12


Definition of TE (2)

 In network management/planning, you can use various


techniques at different time scales:

14 April 2005 - © RUN, ULG 13


Definition of TE (3)

 Two different approaches to TE:


 Strategic TE
 Proactive approach: configure in anticipation of traffic
changes…
 Medium/Long term process.
 Tactical TE
 Reactive approach: configure in response to traffic changes…
 Short term process.
 We are more focused on the first approach.

14 April 2005 - © RUN, ULG 14


Definition of TE (4)

 Most problems encountered in traffic


engineering are large and combinatorial.
So, efficient and quasi-optimal heuristics
are needed…
 The number of paths from a source to a
destination is O(2n), with n the number of
nodes.
 The number of node pairs is O(n2).

14 April 2005 - © RUN, ULG 15


TE solutions – Introduction

 We now present different solutions to


balance the traffic in the following problem:

A D

Load balancing:
C G
Find a way to balance the
traffic to G on CDG and
CEFG.
B E F

14 April 2005 - © RUN, ULG 16


TE solutions - Pure IP

 First solution: find an appropriate set of weights, i.e. such


that the resulting shortest paths minimise the maximum
link load.

A D Cost of CDG: 6
3 3 Cost of CEFG: 6

C G The cost of the two paths


are equal, so they will be
2 2 used in parallel (if an option
B E F of the routing protocol called
2 ECMP is activated).

 Note that this problem has been proven to be NP-hard (heuristics are needed).

14 April 2005 - © RUN, ULG 17


TE solutions - MPLS

 Another approach is to factorise the


problem into smaller and simpler ones.
 MPLS (Multi Protocol Label Switching) can
help us to do that.

14 April 2005 - © RUN, ULG 20


TE solutions - MPLS (2)

 Thistechnology allows us to establish tunnels


(LSPs, label switched paths) in the topology.

A LSP from A to G
D
We can freely choose the
path and the granularity of
C G each of these LSPs.

These are independent from


B LSP from B to G E F each other.

14 April 2005 - © RUN, ULG 21


TE solutions – MPLS (3)

 We are now able to route all the flows


according to a particular policy (load balancing
for instance).
 In case traffic changes, we can reroute only
some LSPs to adapt the solution. So we are
much more dynamic than with the pure IP
solution.

14 April 2005 - © RUN, ULG 22


TE solutions – MPLS (4)

 In MPLS, there is a protection technique


called Fast Reroute (FRR) which allows to
recover from a failure in tens of
milliseconds. The key idea is that the
protection is local.

14 April 2005 - © RUN, ULG 23


TE solutions - MPLS (5)

A LSP1 from A to G
D

C G If CD or D fails, the traffic


is automatically rerouted by
Detour LSP for LSP1
C on the blue LSP.

B E F

A backup LSP is only active if its primary LSP


is down. As we can freely choose the path of a
backup LSP, we can compute it such that no
congestion appears in case of failure.
14 April 2005 - © RUN, ULG 24
TE solutions - Conclusion

 MPLS routing gives more flexibility than


pure IP routing but is somewhat more
complicated.
 Finally, it is possible to use hybrid
solutions:
 Most of the flows are routed using IP.
 Other flows are routed using MPLS.
 Avoid a full mesh of LSPs (impractical for
large networks).

14 April 2005 - © RUN, ULG 25


Outline

 Introduction to traffic engineering (TE)


 The TOTEM project (TOolbox for Traffic Engineering Methods)
 Objectives
 TE algorithms already integrated
 Architecture
 Case Study
 TE research performed in RUN
 Conclusion

14 April 2005 - © RUN, ULG 26


Outline

 Introduction to traffic engineering (TE)


 The TOTEM project (TOolbox for Traffic Engineering Methods)
 Objectives
 TE algorithms already integrated
 Architecture
 Case Study
 TE research performed in RUN
 Conclusion

14 April 2005 - © RUN, ULG 27


Objectives

 A lot of research has been done


→ good solutions exist

 Only few of them are currently used in real


network

 One reason : these methods are implemented for


research and simulation purpose.

14 April 2005 - © RUN, ULG 28


Objectives (2)

 Main TOTEM objective :


to fill the gap between the academic and
operational worlds by providing interfaces
with existing tools

 The toolbox can also be used by the researchers who are


interested in testing, comparing and promoting their
research.

 Licensed under the LGPL license :


 free download : http://totem.run.montefiore.ulg.ac.be.

14 April 2005 - © RUN, ULG 29


Integration in an operationnal network

Network
Management TOTEM Toolbox
System

Provisionning tool

14 April 2005 - © RUN, ULG 30


Simulation environment

Topology.xml

TOTEM Toolbox

TrafficMatrix.xml

Scenario.xml

14 April 2005 - © RUN, ULG 31


Outline

 Introduction to traffic engineering (TE)


 The TOTEM project (TOolbox for Traffic Engineering Methods)
 Objectives
 TE algorithms already integrated
 Architecture
 Case Study
 TE research performed in RUN
 Conclusion

14 April 2005 - © RUN, ULG 32


TE solutions already integrated

 Intra-Domain
 IGP-WO:
IGP Weight Optimisation
 DAMOTE: path computation algorithm

 Inter-Domain
 C-BGP: an efficient BGP simulator

14 April 2005 - © RUN, ULG 33


Intra-Domain : IGP-WO

Topology
IGP - WO Set of
Weights
Traffic Matrix

 Consider ECMP routing enabled


 Based on Tabu Search Heuristic
 Objective function (to minimise) :
sum over all links of above objective function

14 April 2005 - © RUN, ULG 34


Intra-Domain : DAMOTE

 Path Computation algorithm for LSP (tunnels) in an MPLS network


 Designed to be implemented on routers (decentralised algorithm)
 Based on a modified Bellman Ford algorithm
 Goal : balance the load over all the links of the network :

LSP Request :
- source node Computed Path
DAMOTE
- destination node
- bandwidth

14 April 2005 - © RUN, ULG 35


Inter-Domain : C-BGP

 Simulate the exchange of inter-domain BGP routing


messages (need of BGP messages dump)

 Take into account all the steps of the (complex and


thus not well known !) BGP decision process

 Can compute inter-domain routes (one node is an


AS) by simulation

14 April 2005 - © RUN, ULG 36


Outline

 Introduction to traffic engineering (TE)


 The TOTEM project (TOolbox for Traffic Engineering Methods)
 Objectives
 TE algorithms already integrated
 Architecture
 Case Study
 TE research performed in RUN
 Conclusion

14 April 2005 - © RUN, ULG 37


Architecture

 Software architecture objectives :


 Minimise integration work of a new algorithm
 Provide interoperable interfaces
 Provide different execution modes
 Allow interactions with algorithms written in
different languages

14 April 2005 - © RUN, ULG 38


Architecture (2)

 Developed in Java for :


 Flexibility
 Object oriented abstraction level
 Dozen of libraries to deal with XML, graphs,...
 JNI (Java Native Interface) to integrate algorithms
written in C or C++.
→ (3 main contributions were developed in C and integrated)

14 April 2005 - © RUN, ULG 39


Architecture (3)

14 April 2005 - © RUN, ULG 40


Outline

 Introduction to traffic engineering (TE)


 The TOTEM project (TOolbox for Traffic Engineering Methods)
 Objectives
 TE algorithms already integrated
 Architecture
 Case Study
 TE research performed in RUN
 Conclusion

14 April 2005 - © RUN, ULG 41


Case study : the GÉANT network

 European research
network
 30 countries and 26
national networks
 23 nodes
 38 links

14 April 2005 - © RUN, ULG 42


Traffic analysis

 Objective : compare different routing schemes


 MCNF (Multi Commodity Network flow) whose objective
is to minimize max loaded link.
 CSPF with GÉANT metrics
 CSPFHopCount with 1 as metric for each link
 CSPFInvCap with the inverse of bandwidth as metric
 CSPFInvFreeBw with the inverse of residual bandwidth
as metric
 DAMOTE

14 April 2005 - © RUN, ULG 43


Traffic Analysis

Comparison of different routing schemes

Algorithm Max Per 10 Mean Std dev


MCNF 29,9 % 26,9% 10,4% 9,6%
CSPF 50,5% 18,1% 7,4% 8,4%
CSPFHopCount 83,5% 19,8% 10,2% 15%
CSPFInvCap 43,5% 16,5% 7,1% 7,2%
CSPFInvFreeBW 30,3% 13,7% 7,1% 6,2%
DAMOTE 30,3% 14,9% 8,9% 5,3%

14 April 2005 - © RUN, ULG 44


Outline

 Introduction to traffic engineering (TE)


 The TOTEM project
 TE research performed in RUN
 Decentralised Agent for MPLS Online Traffic Engineering
(DAMOTE)
 Conclusion

14 April 2005 - © RUN, ULG 45


RSVP-TE

PATH

RESV

14 April 2005 - © RUN, ULG 46


DAMOTE

Backup path computation algorithm


 Fast restoration with local backup LSP
 Try to minimize the bandwidth reservation for backup path


Local restoration requires a lot of bandwidth reservation


Solution : Share the bandwidth between backup paths that
are not supposed to fail at the same time.

14 April 2005 - © RUN, ULG 47


DAMOTE

Backup - backup sharing

14 April 2005 - © RUN, ULG 48


DAMOTE

Primary - backup sharing

14 April 2005 - © RUN, ULG 49


DAMOTE


Need a lot of information !

Flooding all the needed information in all the network :
→ not scalable !

Our solution : To keep in memory all the information that it
is possible, i.e.
- From the computation of backup paths
- From the RSVP-TE message

In addition to this, only a few information has to be flooded
in the whole network.

Key point : each node compute all the backup paths
protecting it and incoming links.

14 April 2005 - © RUN, ULG 50


DAMOTE

Information Flow

14 April 2005 - © RUN, ULG 51


Outline

 Introduction to traffic engineering (TE)


 The TOTEM project
 TE research performed in RUN
 Conclusion

14 April 2005 - © RUN, ULG 52


Conclusion

 Future works :
 Study hybrid IP / MPLS methods
 Combine intra/inter-domain TE
 Dynamic adaptation to network conditions
→ more robust methods
 Deployment of the toolbox in a real testbed
network (with University of Napoli)

14 April 2005 - © RUN, ULG 53


About RUN team...

 1 professor – Guy Leduc


 9 researchers – 1 visiting researcher
 Research topics :
 Trafficengineering
 Mobile network
 Ad-Hoc network
 TCP congestion control
 Active and P2P network
 Multimedia transmission

14 April 2005 - © RUN, ULG 54


Traffic engineering project (1)

 TOTEM : Toolbox for Traffic Engineering Methods


 Funded by the DGTRE – walloon project
 3 years : Nov 04 – Oct 06
 3 teams from UCL and ULG
 10 researchers and 3 professors
 35 papers and 9 IETF draft

 E-Next : European Network of Excellence


 Jan 04 – Dec 05
 Leader of the task force 4.2 on traffic engineering
 Collaboration with Belgium, Italy, Spain, Sweden, Finland,
Austria, France, ...

14 April 2005 - © RUN, ULG 55


Traffic engineering project (2)

 Integration and test of traffic engineering methods for IP/MPLS


networks
 Funded by the CGRI – French community
 2 years – Jan 05 – Dec 06
 Collaboration with Napoli
 GRAAL : GRAphs and ALgorithms in communication networks
 European COST program (Sep 04 – Oct 08)
 Former project
 Atrium : A Testbed of Terabit IP Routers Running MPLS over
DWDM
 Funded by European Commission in the IST framework

 3 years : Jan 2001 – Dec 2003

14 April 2005 - © RUN, ULG 56


The end

That's all folks !

Thank you for your attention.

14 April 2005 - © RUN, ULG 57

You might also like