You are on page 1of 4

CS4450 Internetworking

Based on

Content Source Materials

Internet Traffic Engineering


Dr Chandana Gamage Department of computer Science and Engineering University of Moratuwa

June 2010

Presentation Interior Gateway Protocol (IGP) Metric Based Traffic Engineering by Visa Holopainen, Networking Laboratory, Helsinki University of Technology, Finland Papers in IEEE Communications Magazine, Oct. 2002, Vol. 40, No. 10, pp 110-131 by N Brownlee and K C Claffy, Understanding Internet traffic streams: dragonflies and tortoises. B Fortz, J Rexford and M Thorup, Traffic engineering with traditional IP routing protocols. R M Mortier, Multi-timescale Internet traffic engineering. Presentation Internet Traffic Engineering by Jeng-Long Chiang Technical Report Internet Traffic Engineering by Richard Mortier, Computer Laboratory, University of Cambridge
2

Concept of Traffic Engineering (TE)

Concept of TE

Traffic Engineering (TE) is a field of communications engineering that tries to make network operations more effective and reliable while at the same time optimizing resource utilization Application of technology and scientific principles to the
Measurement, Characterization, Modeling, and Control

Traffic engineering involves adapting the routing of traffic to network conditions, with the joint goals of good user performance and efficient use of network resources Traffic analysis helps the operators on arranging network resources and applying pricing policies

of Internet traffic

Internet Traffic Streams

Problems for TE to Solve


Internet traffic streams can be classified by size and lifetime Example short streams
By size: small v.s. high-volume stream By lifetime: short v.s. long-run stream

Non-Web TCP = 30B, 50B, 800B, 1500B (<3KB) UDP = 30B, 80B (<300B)

Web TCP = 20B, 50B, 100B, 200B (<300B, failed request) 300B-800B (<40KB)

Effective bandwidth utilization within an Autonomous System (Intra-AS) Optimal policy usage between Autonomous Systems (InterAS / BGP TE) Fast connectivity restoration after a component breakdown (IGP Fast Convergence, MPLS Fast Re-Route, etc.)

A TE Goal - Better Throughput

A TE Goal - More Equal Link Utilization

Click to add an outline


Without TE
90 80 70 60

Click to add an outline


Without TE
90 80 70 60 50 40 30 20 10 0 10 20 30 40 50 60 10 20 30 40 50 60 Link 1 Link 2 Link 3

With TE

With TE

Throughput (Mbit/s)

Throughput (Mbit/s)

70 60 Src-Dst 1 50 40 30 20 10 0 10 20 30 40 50 60 Src-Dst 2 Src-Dst 3 Average

70

Utilization (%)

60 50 40 30 20 Link 1 Link 2 Link 3

40 30 20 10 0 10 20 30 40 50 60

Src-Dst 1 Src-Dst 2 Src-Dst 3 Average

10 0

T ime (min)

Time (min)

T ime (min)

Utilization (%)

50

T ime (min)

Requirements for TE: Policy System


TE Method Classification

TE systems consist of a set of rules (Policy) that are propagated to enforcement points Hierarchical policy systems possible Business policy: Utility function policy: Action policy:
f(High responsiveness) Response time must be < 2s f(Response time must be < 2s) condition C if condition C then do action A

TE methods can be classified in many ways:

Short term vs. Long term TE Intra-domain vs. Inter-domain TE Centralized vs. Distributed TE On-line vs. Off-line TE Performance optimizing vs. Availability maximizing TE Host-based vs. Network-based TE

10

Example TE Method 1: ECMP TE


ECMP TE (contd.)

Equal Cost Multipath (ECMP) balances load for all equal cost paths towards a destination Every router performs a hash on received packets to determine which one of the paths should be used for the packet The hash may be a function of source address, destination address, source port, destination port etc. and can be static or dynamic The hashing can be performed separately for each packet or based on flows Medium term, Intra-domain, Distributed, (Off-line), Network-based, Availability maximizing
11

ECMP is often considered to be sufficient TE method if both topology and traffic demands are symmetrical like in the figure

Often not enough for complicated networks


12

Example TE Method 2: MPLS TE

Example TE Method 3: IGP Metric Based TE

MPLS TE enables explicit (source) routing, which is quite significant, since it makes unequal-cost load sharing possible In the figure, traffic load is divided evenly to unequal-cost router-paths xyz and xkvz

Tactical IGP Metric Based TE tries to alter link costs in some specific points of congestion
Often only shifts the point of congestion

MPLS TE is somewhat complicated to deploy but when correctly configured, it is very effective
13 14

IGP Metric Based TE (contd.)

IGP Metric Based TE (contd.)

Strategic IGP Metric-Based TE tries to find an optimal link weight setting for the network (using some optimization goal(s))

The main idea in strategic IGP Metric Based TE is to

(1) obtain topology and traffic information from the network, (2) feed this data to an optimization algorithm, and (3) send the optimal link weights back to the network

15

16

IGP Metric Based TE (contd.)

Optimization Tool (IGP-WO)

The routers calculate routes based on the optimal weights and traffic gets forwarded to optimal paths

Interior Gateway Protocol - Weight Optimizer (IGP-WO) is based on a tabu-search heuristic and developed at UCL, Belgium Tries to find a link (interface) cost setting that balances load in the network Uses iterative search and a utility function to discover nearoptimal interface costs for SPF routing Iterative search methods have the risk of visiting old solutions again (cycling) Tabu-search meta-heuristic (a mathematical optimization method using local search techniques) is used to prevent this
18

17

Conclusions

Streams in the Internet continue to become larger TE can be treated as a network operations task, rather than the responsibility of the underlying routing protocol Traditional shortest path routing protocols are surprisingly effective for engineering the flow of traffic in large IP networks Multi-timescale (data, control and management timescales) TE techniques allow dealing with the competing desires of the operators to simplify the services they offer while still providing sufficient flexibility for users to express their individual requirements
19