1 MPLS In Perspective Kireeti Kompella Distinguished Engineer Juniper Networks

1

MPLS In Perspective

Kireeti KompellaDistinguished Engineer

Juniper Networks

Copyright Juniper Networks,

2001 2

Menu

IP salad with horseradish dressingATM flambeeMPLS stewed in its own juice

For aftersServices double espressoRevenue a la mode


2001 3

IP – Good Enough ™

Well-architected, worked out in detail – NOT!Realization: can’t predict the futureMake it reasonableMake it flexibleMake it extensible

stuff above

transport

network

stuff below


2001 4

So Easy to Forget

IP started out with e-mail …… and data services

ftpnews

Now: the “Web”, voice, video, …Also, SLAs, grades of service, …


2001 5

IP Control Plane

Again, just good enoughBut again, flexible, extensible

DV routing was fine for quite a whileJust in time, along came link stateNow: is convergence “in a few seconds” good enough?


2001 6

Good Enough™ Can Get Better

Fast to ultrafast convergence“Bullet-proof” IP

Hitless restart?

“Business” IPMake me money – new services, GoSDon’t lose me money – uptime, SLAs


2001 7

ATM – Perfectionist’s Dream

Connection-orientedDoes everything and does it wellAnticipated all future uses and factored them inPhilosophical mismatch with IP

stuff above

transport

network

ATM

AAL

1AA

L 2

AAL

3/4

AAL

5


2001 8

MPLS

If (ATM = Frame Relay on steroids)then (MPLS = ATM on happy juice)

Make it just Good Enough ™Despite all efforts to make it perfect

IP control planeIP philosophy


2001 9

What Does MPLS Offer?

TunnelsDrop a packet in, and out it comes at the other end without being IP routed

Explicit (source) routing (circuits)Label stack

2-label stack: “outer” label defines the tunnel; “inner” label demultiplexes

Layer 2 independence


2001 10

Why Tunnels?

Can’t IP routeNon-IP packetsIP packets with private addresses

Don’t want to IP route“BGP-free” coreMulticast


2001 11

How Tunnels?

MPLS: LDP – “automagic” tunnels that follow IP routingIP: IP-in-IP, GRE, IPSec, UTICan one tunnel do multiple things?

Tunnel demux


2001 12

Tunnel Comparison

MPLS (LDP) tunnelsSmall headerLabel stackingSignaling for demuxAutomagic tunnelsTracks IP routingHarder to spoofNo data security

IP tunnelsBig headerNo stacking (*)No signaling (yet)Configured tunnelsDuh!SpoofableIPSec


2001 13

Bottom Line on Tunnels

Don’t need MPLS for tunnelsBut MPLS tunnels have some nice propertiesDecision (should be) based on cost of deploying new protocol vs. benefits


2001 14

Why Explicit Routing?

Traffic EngineeringFast rerouteGuaranteed bandwidthProbably othersConnection-oriented paradigm nicely complements IP’s connectionlessness


2001 15

Traffic Engineering

Is ATM the best way to engineer traffic?Or is it MPLS?Or can we do just fine with IP?

First question: do you need traffic engineering? What part of network?


2001 16

Traffic Engineering Steps

First, determine how to lay out traffic on the physical topology

Measure traffic (e.g., city-pair-wise)Crunch numbers

Second, do something to convince the packets to follow your plan


2001 17

Traffic Engineering Options

BGP – play with communities, filteringIGP – play with metrics

Linear programming can help

Source routingATMMPLS


2001 18

Traffic Engineering

Warning: read at your own risk!Fine-grained Traffic Engineering needs some form of source routingSpecific incremental changes much easier with source routing

Change a single city-pair flowReacting to a link failure


2001 19

Linear Programming

TE among N cities: N² city pairsSet up N² by N² matrix for LPMatrix multiplication/inversion is O(M³) for M x M matrix; simplex is O(M³) matrix operationsSo, LP problem is O(N12)Also can’t deal with “looped routes”


2001 20

Fast Reroute

Can MPLS re-route as fast as SONET (50ms)?Can IP re-route as fast as MPLS?Do packets get dizzy if they are re-routed too fast?


2001 21

Fast Reroute (2)

First question: how fast is fast?Do you really need 50 ms failover?

Second question: can you reroute really quickly while maintaining network stability?Third question: what are the scalability issues with fast reroute?


2001 22

Fast Reroute Comparison

IPAll nodes must be told of failure

Fast propagation, fast SPF trigger: how stable?One step to full reconvergence

MPLS (RSVP-TE)Only the two ends of the link need be told (no signaling)Local operation: explicit routing; more stableTwo step process: detour + converge


2001 23

Fast Reroute: MPLS vs. IP

A B

C

100010

10

IP routing to B

pkt to B

MPLS detour to B


2001 24

Guaranteed Bandwidth

Again, first question: do you need it?If so, you need source routing, CAC and some way of signaling b/wRSVP-TE can do thisATM could probably do it better


2001 25

“MPLS” Services

IP VPNs (RFC 2547 et al)Layer 2 transportLayer 2 VPNsTransparent LAN ServiceTDM over MPLS over TDM over …Electricity over photons?Have we gotten a little carried away?


2001 26

“MPLS” Services (2)

Most of these services need tunnelsNot really MPLS services

MPLS-geeks definitely responsibleSome of these services enhanced by source routingMore services may mean more revenue, could also keep you awake at night


2001 27

Revenue

RFC 2547New service – recent deploymentGive it a shot, or run like hell? Or wait?

Layer 2 VPNsOld service – lots of deploymentNew transport – is it Good Enough?

Guaranteed bandwidth, Diff Serv, …?


2001 28

Things to Ponder

Can Good Enough™ IP stay ahead of the curve?Even if so, can MPLS help?

Is MPLS a support, a crutch or a banana peel?Is connection-orientedness a useful addition to connectionless IP?

What services, when, how far to go?


2001 29

My Biases

VendorMPLS geekProtocols freakNeutral about ATMIP rules!Reasonably agnostic

30

http://www.juniper.net

Thank you!

[email protected]

Documents

1 MPLS In Perspective Kireeti Kompella Distinguished Engineer Juniper Networks