Transcript P00555: Multiservice Networks

MPLS
Topics:
Introduction to MPLS
Tutorial Questions and Recommended Reading
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-1
Routing Without MPLS

Large organizational networks are heavily subnetted for structuring
purposes
◦ Packets crossing such a network could transit many routers

Connectionless nature of IP poses two challenges
◦ Per-packet processing
◦ Distributed routing (Per hop behaviour PHB)

Core routers needs to forward tens of millions of packets per
second
◦ Must process each packet in a few nanoseconds
 Potential for congestion in router forwarder (packet switch)

Packets follow the best path according to the routing table in each
router
◦ No opportunity for setting end-to-end path
◦ Could override with static routes
 But this approach doesn’t scale
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-2
Normal IP Transit Network
BGP Routers
to Service Provider
Enterprise Network
Service Provider’s
Transit Network
Enterprise Network
Normal
Subnet Routers

Connecting widely separated parts of an enterprise network
◦ ‘Enterprise’: a large (national, international) company, organization
◦ a number of sites (campuses, branches, offices)
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-3
IP vs. MPLS Transit Networks
Service Provider’s
Transit Network
IP
Enterprise Network
Normal
IP routers
Enterprise Network
Routed Path
MPLS
MPLS transit network
MPLS-enabled
routers
Enterprise network
(routed normally)
Label Switched Path
Packets and Circuits: Chris Cooper Feb 2005
Enterprise network
(routed normally)
P00555-3-4
Benefits of MPLS

Lower transit delay
◦ Core routers switch not route

Traffic engineering
◦ Packets can take non-standard
path(OSPF path)

Scalability
◦ Labels can be nested
to facilitate network
hierarchy

Flexibility
◦ Can be used over
 LANs
 PPP tunnels
 [ATM & Frame Relay backbones]
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-5
MPLS Goal: Dynamic Switched Flows

Original goal: bypass routing table lookup for as many packets as
possible
◦ Dynamically detect packet flows
 Identified by unique pairs of IP addresses and port numbers
◦ Switch, rather than route, packets on known flows
 Cisco called this “route once, switch many”

Original approach based on two TCP-related assumptions
◦ Majority of IP packets belong to TCP sessions
 Rather than UDP datagram streams
◦ TCP sessions have (relatively) long duration
 File transfers, conferencing

Increasing popularity of Web browsing undermined this goal
◦ Uses short-duration sessions
 Per-flow path setup doesn’t scale
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-6
MPLS Goal: Dynamic Switched Flows
(continued)
MPLS transit network
MPLS-enabled
routers
Enterprise network
(routed normally)
Enterprise network
(routed normally)
Packet flow
following switched path
Flow-detecting MPLS routers
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-7
MPLS Goal: Traffic Engineering

Determine end-to-end path for given packet flows
◦ Override routing protocol decision where administratively appropriate

Allows routing policy to be set
◦ Reflect service offerings
 Low-delay path for voice traffic
 More secure path for certain customers

Now seen as most important reason for using MPLS
MPLS transit network
MPLS routers
Predetermined path
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-8
Multiprotocol Label Switching

Overview

Label Switching

Operation
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-9
Label Switching

Labels packets for faster switching through network
◦ Connection-oriented protocols use virtual circuit ID
 Frame relay DLCI
 ATM VPI/VCI
◦ Connectionless protocols need to add label
 VLAN identifier (802.1Q trunking)
 MPLS label (added to Ethernet and PPP)

Switches set up paths as required
◦ Associate labels with paths
◦ Use label as route-table lookup

Labels often have only link-by-link significance
◦ Allows switch to differentiate incoming flows
◦ Each switch maps label values predictably for outgoing flows
DLCI = data-link connection identifier
VCI = virtual channel identifier
VPI = virtual path identifier
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-10
Label Switching Routers

MPLS-enabled devices are called Label Switching Routers (LSRs)
◦ Multilayer switches enhanced with MPLS protocols

MPLS identifies two roles for LSRs
Edge LSR
◦ Edge LSRs and Core LSRs
 Edge LSRs often called
Label Edge Routers (LERs)

Edge LSRs
Edge LSR
Core LSRs
◦ Determine packet path and perform flow classification
◦ Assign unique labels to each flow

Core LSRs
◦ Use label values to switch packets over cut-through paths
◦ Layer 2 forwarding bypasses normal routing function
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-11
Multiprotocol Label Switching

Overview

Label Switching

Operation
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-12
Operational Overview I

Identify groups of packets of packets travelling over a
common path
◦ For example, towards the same destination network or
host
◦ Called a forwarding equivalence class

Assume they have common forwarding requirements
and assign a label to each group
◦ Encapsulate with label header carrying same label value
◦ Communicate label settings to downstream router
◦ Downstream router assigns label to outgoing FEC and
communicates downstream
 And so on
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-13
Operational Overview II

Once LSP has been set up routers process FEC accordingly

Ingress LSR (ingress LER) adds label to packet

Core LSRs match incoming labels to route table, which gives
output port
◦ Outgoing label map applies downstream label value
 As previously communicated to downstream router
◦ Bypassing conventional packet-by-packet, hop-by-hop L3 processing

Egress LSR (egress LER) removes it

Set of label mappings for a group constitutes the label switched
path (LSP) for that FEC
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-14
Label Switching
b
c
b
c
p
s
q
r
5
1
x
b
x
b
6
2
y
y
z
z
m
m
7
3
a
a
s
s
8
4
Port 1
route table
Incoming
label
Outgoing
port
Incoming
port
Incoming
label
Outgoing
label
a
8
4
a
p
b
5
1
b
q
c
5
1
c
r
d
7
3
z
s
Packets and Circuits: Chris Cooper Feb 2005
Port 5
label map
P00555-3-15
Label Switched Path (LSP)

MPLS identifies two types of label switched path
◦ Destination- based(follow the OSPF path)
◦ Explicitly routed( follow the way u determined)
 (Cisco terminology; descriptive)

Destination-based LSP follows conventional forwarding path
◦ As determined by IP routing table
◦ Originally set up from destination LER source LER
 (Why is that?)

Explicitly routed LSP use source-specified path (source routing)
◦ Path set up from source LER destination LER
◦ Useful for overriding normal route selection based on least cost path
 E.g. for enforcing route selection (‘routing policy’)
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-16
LSP Set-Up

LSR is essentially an MPLS-enabled multi-layer switch

Routing database constructed in normal way

Switching engine assigns labels to forwarding paths

Once LSP set up, packets are label-switched(packets can flow)

Conventional, in most cases, for router prior to egress LSR to
remove label
◦ Using, for example, OSPF
◦ Then made available to MPLS switching engine
◦ Sends route/label mappings to next-hop neighbour using a/the Label
Distribution Protocol (LDP)( the path are renewed periodically)
◦ Avoids processing load on LSR
◦ Called penultimate hop-popping
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-17
How MPLS Works
(continued)
Normal
IP packet
Normal
IP packet
Labelled packet
May pop label
Edge LSR
Edge LSR
Core LSRs
Label Switching Path (LSP)
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-18
MPLS Header

Header is 32 bits (4 octets)
◦ Label field is 20 bits
◦ Three-bit EXPerimental field can be used to carry (some) DiffServ
markings through MPLS network
◦ S = 1 indicates bottom of stack
◦ Time To Live (TTL) is decremented by LSRs to maintain usual
packet hop count
Number of bits
20
Label
3
1
EXP S
8
TTL
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-19
Label Encapsulation Schemes

Format: 32 bits added to beginning
of packet (prepended)
◦ 20-bit label
◦ Remaining bits used for variety of
purposes

C
R
C
IP payload
Two ways of inserting label
Label inserted as additional header
◦ Add as protocol shim to existing
header
 PPP and switched LANs
 usual method
IP payload
◦ Map onto existing virtual circuit
identifier
 ATM or Frame Relay
 rare: included for completeness

LAN/PPP
header
Once the label is in place,
established schemes can be used
◦ E.g. label multiplexing and switching
C
R
C
Label in FR DLCI field
IP partpayload
Label in ATM VPI/VCI fields
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-20
Label Stacking I

Can add further label in front of (‘on top of’) the existing one(s)
◦ Nested labels treated as stack
 Hence use of term ‘pop’
◦ Network switches on top-most label

Allows several LSPs to be grouped for forwarding purposes
◦ Provided they can be treated as a single FEC
 E.g. all heading to same edge-point

Can continue the process, grouping groups together into a further group,
with a new label

Each LSP marked by pair of label edge routers and a label in the stack
◦ Ingress LER pushes new label onto stack
◦ Egress LER pops label off the stack
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-21
Label Stacking II
LERs
LSP
LSP
LERs
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-22
MPLS Summary I

Multiprotocol Label Switching (MPLS)
◦ Applies label switching to IP networks
◦ Facilitates
 Unequal cost load balancing
 Setting routing policies (traffic engineering)
 Virtual Private Networks (VPNs)
◦ Bypasses potential bottlenecks causes by large route table look-ups
◦ Allows provider network nesting though label stacking


Allows label mapping to be communicated in variety of ways
How do u communicate label route information across the network
◦ LDP(label distribution protocol)
◦ OSPF and BGP enhancements
◦ RSVP


Details of how to recover from link failure still being finalized
Generalised MPLS: paths over SONET/SDH & wavelengths (‘s’) in WDM
networks
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-23
Tutorial topics: MPLS
1.
What are the similarities between MPLS label
switching and FR/ATM switching?
2.
What (if any) are the differences?
3.
Look up “penultimate hop popping”. What is it
and what does it achieve?
4.
Why is a ‘destination-based’ MPLS path set up
from destination LSR back towards source LSR?
1. Hint Remember path is unidirectional: think about label mapping
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-24
STUDY QUESTION
Q1(i) What does the term label switching mean?
Explain, using appropriate diagrams, how MPLS
applies this principle to IP packets and the main
advantages it gives. [5]
(ii) Figure 1 shows part of an OSPF network. The
network administrator notices that traffic from
the remote site LANs frequently congests the
route to Head Office. Explain why this is, and
describe, with examples, how MPLS could be
used to overcome this problem. [5]
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-25
FIGURE 1
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-26
Q 2. Explain what is meant by a Forwarding Equivalence
Class and a Label Switched Path in MPLS. What is
meant by 'DiffServ-enabled MPLS'? Explain briefly what
MPLS labelling arrangements you would expect to see
for a set of enterprise VPNs, where each VPN supports
its own two DiffServ per-hop behaviours (PHBs). Would
this change, and if so, how, if within each VPN, an
enterprise also used MPLS to traffic engineer the
routing of traffic with a different per-hop behaviour?
Explain your answer. [6]
Packets and Circuits: Chris Cooper Feb 2005
P00555-3-27