MPLS Traffic EngineeringB1%E8%C3%A2... · 2012-05-06 · 2 Internet Architecture Team / ETRI KRnet 2001 Table of Contents What is MPLS-TE? Definition, Necessity, Goals, and Vocabulary

MPLS Traffic Engineering:Checking Up Real-world Applicability

한국전자통신연구원 (ETRI)인터넷기술연구부

김창훈 {[email protected]}

KRnet 2001Internet Architecture Team / ETRI2

Table of ContentsWhat is MPLS-TE?

Definition, Necessity, Goals, and Vocabulary

MPLS-TE MechanismsLSP Routing, Traffic Protection, Load Balancing & Sharing, Path Reoptimization, FA LSPs, etc.

Real-world Deployment IssuesMPLS-TE Deployment Scenarios Practical Issues on MPLS TE DeploymentTE Servers

Ending Remarks



Definition, Necessity, Goals, and Applications


Real-world Deployment IssuesMPLS-TE Deployment ScenariosPractical Issues on MPLS TE DeploymentTE Servers

Ending Remarks


What is Traffic Engineering?

The task of mapping traffic flows onto an existing physical topology to facilitate efficient and reliable network operations

Check mpls & tewg working group documents for more well-versed definitions

Requirements for Traffic Engineering Over MPLS (RFC 2702)A Framework for Internet Traffic Engineering (draft-ietf-tewg-framework-05.txt)


Legacy Internet TE EffortsIGP Metric-Based TE

Remember “fish problem?”Drawbacks

“Blame Shifting”: only serves to move problem aroundLacks granularityInstability

Overlay Network ApproachATM core ringed by routers & overlaid PVCs on top of itDrawbacks

Full mesh overheadNot well integratedCell TaxATM SAR speed


Benefits of MPLS

Low-overhead VCs for IPOriginally designed to make routers fasterNot true anymore!

Value of MPLS is now in TE, becauseMPLS can provide different path with different attributes for each different traffic flowIt doesn’t mean that metric-based TE can not accomplish this goal at all

a moot point!


MPLS-TE!

Advantagesthe physical path of the “traffic-engineered path” is not limited to what the IGP would choose as the shortest path to reach the destinationprovides stand-by secondary paths and precomputed detouring pathsprovides strongly unified measurement and control for each “traffic-engineered path”provides traffic aggregation and disaggregation


VocabularyLSP (Label Switched Path)

the “traffic-engineered path”

Primary and Secondary Pathsan LSP can contain a primary path & zero or more secondary paths

Named Patha sequence of explicit hops

LSP A

Primary Path Secondary Path

LSP B

Primary Path Secondary Path

Named Path 1

Named Path 2


Vocabulary – cont’d

Traffic Trunk (TT)an aggregation of traffic flows going from an ingress to an egressforwarded through a common path with common TE requirementscharacterized by

its ingress and egressFEC a set of attributes that determines its behavioral characteristics


Vocabulary – cont’d

Types of LSPsStatic LSPsLDP signaled LSPsRSVP/CR-LDP signaled LSPs

Explicit-path LSPsConstrained-path LSPsNote: both of the two above are not mutually exclusive!


Components of MPLS-TE

Packet Forwarding ComponentMPLS, label switching itself

Information Distribution ComponentIGP (OSPF/IS-IS) extension

Path Selection ComponentConstrained Shortest Path First (CSPF) algorithm

Signaling ComponentLDP, CR-LDP, and RSVP-TE

Not all of these required!


How everything fits into?

Link attributes

Link attributesmodification

RSVP signalingTED

LSP pathsCSPFLSP attributes

Routing table

advertised byIGP-extension

operatorinput

computes

structured as

reservation LSPestablishment

topology &resources

advertised byIGP-extension






Ending Remarks


MPLS-TE Mechanisms

LSP Routingwith TE attributes (LSP & Link attributes)dynamic vs. explicit

Traffic Protection (Resilience)secondary paths and fast reroute

Path Reoptimization (Adaptivity)Load Sharing and Balancing

LSP-level traffic bifurcation

LSP Hierarchyforwarding adjacency LSPs, unnumbered links


Configurable LSP AttributesCan specify the following attributes either for each LSP or for each path belonging to the LSP

bandwidth (traffic profile in CR-LDP)constrained (dynamic) vs. explicit pathaffinityadaptivity

reoptimize-timer, reoptimize-eventresilience

(stand by) secondary paths, fast reroutepriority & preemption

setup, holdroute recordhop-limit, cos, etc.


Traffic Protection:Secondary Paths

LSPs may have zero or more secondary pathsSecondary Path Usage Procedure

fault detection failure notification secondary path computation secondary path signaling traffic switching-over ( retry to convert back)

When the path is “stand-by”fault detection failure notification traffic switching-over

Deficitsome loss may still exist


Traffic Protection:Fast Reroute

Automatically reroutes traffic on an LSP using a pre-computed and pre-established detouring pathTraffic Protection Procedure

fault detection enabling detour failure notification (secondary path computation secondary path signaling) traffic switching-over

Failure onrouter C orlink B-C

router A

B

C

D

E

F

detour to skip A-B detour to skip C-D

detour to skip D-Edetour to skip B-C

detour to skip E-F

router A

B

C

D

E

F


Path Reoptimization

Reoptimizationmeans adaptivity to network state evolution or degenerationis not related to fail-over (resilience);

a new path is always computed when topology failures occur that disrupt an established path

disabled by default due to stability concern


Load Sharing & Balancing

Load SharingPath-level Load Sharing

not supported!LSP-level Load Sharing

supported, only when there are multiple LSPs in between the same ingress and egress pairload sharing factor might be manually setup or automatically determined in proportion to the bandwidth of each LSPprefix hashing not accurate!

Load Balancingis implicitly taken into account by CSPF


LSP Hierarchy

An LSP might be used as an IGP link , a.k.a. forwarding adjacency (FA)

legacy hop-by-hop routed packets may use the LSP as a newly added linkother LSPs maybe recursively signaled over the LSP

by LDP, CR-LDP, and/or RSVP-TE

Such LSPs are calledFA LSPsIGP shortcutsunnumbered links


LSP Hierarchy –cont’d

Use of FA LSPLocal use

Only the ingress of the LSP can utilize the LSPGlobal use

Other routers in the same domain can also use the LSPThe ingress must advertise the FA LSP to the IGP as a unidirectional point-to-point linkWe might need to configure a reverse LSP so that the SPF bidirectionality check succeeds

CautionIGP peering over FA LSPs must not be allowed!


Advertising FA LSP and Its Metric

Now, how do we decide a path from A to D?Hop-by-hop A-B-C-D or A-E-(FA LSP)-D

The answer is comparing the metrics!FA LSPs must have static or dynamic metrics

What if both the two metrics are equal?ECMP load sharing

router A B C D

E F

FA LSP E-D

10 10 10

10

20

10 15






Ending Remarks


General Guide for Deployment

High-level Steps for Fast DeploymentIdentify the Problem

congestion, too low or too high utilization, link/node failure, etc.

Gain Lab Experienceon a testbed

Test LSPs in Live Networkon a portion of production network

Deploy LSPs Along IGP Paths and Measureuses only RSVP signaling without CSPF

Reroute LSP to Explicit PathMonitor Results


Deployment ScenariosFuturistic or Ideal Scenario

1. Identify each TT (Traffic Trunk) and its TE requirementsby autonomous policies or by SLA negotiationGranularity of TT might be various, from micro to macro

2. Construct FECs for the TT3. Design an LSP or a set of LSPs to route the TT4. Enforce the LSPs5. Measure and monitor network and traffic behavior6. Periodically adjust the attributes of the TT or the LSPs

Sometime, need to aggregated or disaggregate the TTs and LSPs

Standard information modeling for MPLS TEwork in progress in policy and mpls working groups


Deployment Scenario – cont’d

Practical Scenario (currently feasible)1. Identify a specific TE problem2. Identify contributing flows

LSPs serve best for this purpose also3. Devise a cure for the problem using MPLS TE mechanisms

Adding a new FA LSP vs. rerouting an already existing oneUsing the LSP locally vs. globally?

4. Enforce the TE mechanisms5. Measure and monitor network and traffic behavior(Go to the step 1.)

AdvantagesLegacy hop-by-hop routing and IGP SPF still worksincorporates with metric-based TE


Practical Scenario Example

1. Identify the problem the congestion2.1 Set up LSPs that simply follow the IGP path from

some routers to J2.2 Measure the traffic on the LSPs

LSP A-J : 70Mb the biggest contributorLSP B-J : 20MbLSP C-J : 30Mb, etc.

A

B

CD

E

F

31

1

J

H

G I1

1 1 1

1

1

1

2

1


Practical Scenario Example – cont’d

3. Devise solutions and choose onesolution 1 : redirecting LSP A-J along another pathsolution 2 : adding a FA LSP between E and J that is actually routed along E-F-J

4. Enforce5. Measure and monitor network and traffic behavior

A

B

CD

E

F J

H

G I

1


Deployment Sample Case

Global Crossing ltd.One of the 10 largest ISPs in the U.S.National backbone consists of ~50 POPs of more than 300 routersEnhanced IS-IS used as an IGP (all routers in the same level)All routers also runs BGP

Nation-wide MPLS deployment aimsBetter TE and VPN service provisioning


Global Crossing’s Architecture

Two-tier ModelThe nation-wide network is divided into 9 regionsAll routers in each region are fully meshed to form the 1st layer of LSPs (100 ~ 2,500)Core routers in all regions are also fully meshed to form the 2nd layer of LSPs (100 ~ 2,500)


Global Crossing’s General Guide

ProcedureCollect statistics using MPLS LSPs

By means of deploying LSPs without specifying their BW requirement and explicit pathPurpose is to collect traffic statistics in the interested areaCan induce the traffic matrix for the target network

Deploy LSPs with proper constraintsbandwidthexplicit hopsother TE attributes, if required

Periodic Update of the constraintsOffline Constraint-based Routing


Practical Issues on Deployment

InteroperabilityDeficit of Online Path CalculationProblems regarding to Traffic TrunksMeasurement and Control Issues


Interoperability

Vendor specific implementation details diverge!

Almost everything but signaling standard might be differentUsing more than two heterogeneous families in a domain may cause unpredictable operational problems

Need a unified abstraction system to hide, moderate, and arbitrate the differences


Deficit of Online Path Calc.

Online path calc. considers one LSP at a time

undeterministicThe order in which an LSP is calculated plays a critical role!

Global optimization requiredOptimization tools that simultaneously examine each link’s resource constraints and the requirements of each LSPs all together are necessary


Problems regarding to TTHow to define traffic trunks?

No standardManual classification

requires TE policiesgranularity and scalability concernpractically, only dest. prefix based classification supportedrequires, so called, “policy routing”

BGP-based classificationTransit traffic whose route updates’ next_hop is identical to the egress of an LSP are routed over the LSPonly supported by some implementation

Implicit classification by IGP


Problems regarding to TT – cont’d

How to map a traffic trunk’s attributes onto LSPs’ constraints?

need a global viewmust be able to anticipate the effect, to some extentmust be able to rationalize

by simulationsby measurementsby policiesby intuition?by experience?


Measurement and Control

Measurementprovides rationale and fundamental bases to induce proper TE constraints for TTs and LSPs

such as, traffic (demand) matrices, congestion indication, LSP statistics, etc.

methodsSNMP (various MIBs), CLI, Cisco Netflow and TMS, and/or JUNOS MPLS Statistics, RTFM probes, etc.

Controlmanages TE policies

policy editing, conflict check, enforcement, withdrawal, etc.customized to service specific policies, such as VPN policies


TE Servers

PerformsProvisioning a Unified Management Environment

better interoperability

TE Policy ManagementNetwork & Traffic Measurement and AnalysesMPLS VPN ManagementSimulations and Analyses

failure simulationglobal optimization, etc.

Capacity Planning


TE Servers:Products Introduction

WANDL, Inc. - MPLSView ®Automated data collection, layout, event collection and filtering (mainly focused on pre-configured LSPs)A quasi real-time view on the configuration of the network, including LSP set-up & state and per-LSP traffic flowPartnership with Cisco and Juniper

Makesystems, Inc. - NetMaker ®Network engineering and simulation tool for IP and MPLSMerged to OPNET Technologies, Inc.

Scalable Network Technologies, Inc. - QualNetOff-line network simulator for MPLS

ETRI - Wise<TE> ®


Ending RemarksTraffic Engineering

the key to provide the advanced services, such asVoIP, VoMPLS, VPN, QoS, etc.

the most essential part in network management

MPLS TEone of the best fit for the Internet TEenlarges and expands the opportunity for better TEstill have a lot of interesting and debatable issues on deployment

adopting on-line/off-line TE serversIs MPLS TE really works better than metric-based TE?best practices for MPLS TEand so forth

trade-off!

Documents

MPLS Traffic EngineeringB1%E8%C3%A2... · 2012-05-06 · 2 Internet Architecture Team / ETRI KRnet 2001 Table of Contents What is MPLS-TE? Definition, Necessity, Goals, and Vocabulary