MPLS (Multi-Protocol
Label Switching)
Eric Donnelly
EEL6785
6/19/03
Overview
Background
History
Components, Definitions
Operation
Performance Measurements
Summary
Background
Standard being developed by IETF (Internet Engineering
Task Force) since 1997
Integrates key features of Layer 2 and 3 technologies w/o
limitation to a particular protocol
Packets labeled and sent through network on paths
rather than hop-to-hop as in IP datagrams
Courtesy of [4]
Basic Idea
MPLS is a hybrid model adopted by IETF to
incorporate best properties in both packet
routing & circuit switching
IP Router
Control:
MPLS
Control:
IP Router
Software
IP Router
Software
Forwarding:
Forwarding:
Longest-match
Lookup
Label Swapping
Courtesy of [1]
ATM Switch
Control:
ATM Forum
Software
Forwarding:
Label Swapping
What about GMPLS?
GMPLS (Generalized Multi-Protocol Label
Switching)
Previously Multi-Protocol Lambda Switching
(another MPLS)
GMPLS is deployed from MPLS (Label)
Apply MPLS control plane techniques to optical
switches and IP routing algorithms to manage
lightpaths in an optical network
We will focus on MPLS in this presentation
History
In Mid-90s, many ISPs migrated from router
based cores to IP-over-ATM, this provided:



Greater Bandwidth
Deterministic forwarding performance
Traffic engineering support
Courtesy of [4]
History (Cont.)
No specific Internet backbone networking equipment available for
ISPs.
Equipment needed to be adapted—ATM best solution at time
However, Continued Internet growth increased stress on ATM
networks:



Bandwidth limitations
20 percent “cell tax”
Designed for different tasks (IP—conncectionless, ATM—connectionoriented)
Late 1996, proprietary multilayer solutions emerged with integrated
ATM switching and IP routing:





IP Switching—Ipsilon/Nokia
Tag Switching—Cisco Systems
Aggregate Route-Based IP Switching (ARIS)—IBM
IP Navigator—Cascade/Ascend/Lucent Technologies
Cell Switching Router (CSR)—Toshiba
--These were all similar technologies, but were NOT interoperable
History (Cont.)
Each multilayer switch ran standard IP routing software
(OSPF, BGP-4)
Different label binding approaches

Data-driven model
Label bindings created when data packets arrive.
Labels created either when first packet in a flow or after a number
of packets in a flow have arrived.
IP Switching and CSR used this technique.

Control-driven model
Label bindings created when control information arrives.
Assigned in response to processing of protocol traffic, control traffic
(such as RSVP), or static configuration.
--Control-driven model used in MPLS!
Note:
OSPF-Open Shortest-Path First
BGP-Border Gateway Protocol
RSVP-Resource Reservation Protocol
MPLS emerges
IETF creates MPLS working group to create
unified standard able to operate on any media
infrastructures (Frame Relay, PPP, SONET), not
just ATM.
Uses Control-driven model.
Defines new standard-based IP signaling and
label distribution protocols, as well as existing
protocol extensions (this supports multivendor
interoperability).
Does not implement any of the ATM forum
signaling or routing protocols (eliminates
coordinating of 2 protocol architectures).
Terminology/Components
FEC (Forwarding Equivalence Class)-Group of
packets sharing the same type of transport.
LSR (Label Switched Router)-Swaps labels on
packets in core of network.
LER (Label Edge Router)-Attach Labels to
packets based on a FEC.
LSP (Label Switch Path)-Path through network
based on a FEC (simplex in nature).
LIB (Label Information Base)- MPLS equivalent
to IP routing table, contains FEC-to-Label
bindings.
MPLS Operation
1a. Routing protocols (e.g. OSPF-TE, IS-IS-TE)
exchange reachability to destination networks
1b. Label Distribution Protocol (LDP)
establishes label mappings to destination
network
4. LER at egress
removes label and
delivers packet
IP
IP
2. Ingress LER receives packet
and “label”s packets
3. LSR forwards
packets using label
swapping
Courtesy of [1]
FECs- Group of packets sharing the same
type of transport
All packets in such a group are
provided the same treatment
en route to the destination.
Can be partitioned by:



Source-to-Destination Address
Application
Bandwidth Requirement
Conventional Routing=>packet
assigned to FEC at each hop
(Layer 3 lookup).
MPLS=>done only once (at
egress).
[9]
LSRs and LERs
The devices used for MPLS can be classified into label edge routers
(LERs) and label switching routers (LSRs).
A LSR is a high-speed router device in the core of an MPLS
network.


Participates in the establishment of LSPs, using the appropriate label
signaling protocol
Does high-speed switching of the data traffic based on the established
paths.
A LER is a device that operates at the edge of the access network
and MPLS network.



Supports multiple ports connected to dissimilar networks (such as frame
relay, ATM, and Ethernet)
Forwards this traffic on to the MPLS network after establishing LSPs,
using the label signaling protocol at the ingress and distributing the
traffic back to the access networks at the egress.
Plays important role in the assignment and removal of labels, as traffic
enters or exits an MPLS network.
Inside a LSR
1. Data Plane
2. Control Plane
NHLFE
Label in
Label out
1400
100
500
101
107
103
LIB
FEC DATA
FEC
192.168.10.1
192.168.10.2
192.168.10.3
Protocol
06
11
06
Port
443
69
80
guaranteed no packet
loss
best efforts
controlled load
FEC
Label in
Label out
192.168.10.1
1400
100
192.168.10.2
500
101
192.168.10.3
107
103
Figures Courtesy of [10]
Labels
The MPLS forwarding component is based
on the label-swapping algorithm.
Label encapsulated in MPLS header,
which is “sandwiched” between the Layer
2 and IP header.
If Layer 2 technology supports labels (ATM
VPI/VCI, Frame Relay DLCI), MPLS label
and header encapsulated in the Layer 2
label field.
Why Label Swap?
Label swapping provides a significant number of operational
benefits when compared to conventional hop-by-hop network layer
routing.
Gives an ISP flexibility in the way that it assigns packets to FECs.






Destination address (like conventional IP routing)
Source address.
Application type.
Point of entry/exit to/from the label-swapping network.
CoS conveyed in the packet header.
Any combination of the above.
ISPs can construct customized LSPs that support specific
application requirements (for instance, VPNs). LSPs can be
designed to:



minimize the number of hops
bandwidth requirements
bypass points of congestion
Offer ISPs precise control over the flow of traffic in their networks.
…For Instance
If network core runs conventional longest-match IP
forwarding:


Data from Host A and B follow path 1 since it is the shortest-path
computed.
With MPLS, network administrator could split traffic:
Host A traffic over path 1
Host B traffic over path 2
Courtesy of [4]
MPLS header
Label field- Actual MPLS label (20bits).
CoS field- “Class of Service” can effect queuing and
discard algorithms applied to packets (3 bits).
S (Stack) field- supports a hierarchical label stack (1 bit).
TTL field- “Time-to-live” provides conventional IP TTL
functionality (8 bits).
Courtesy of [4]
…In ATM
…Frame Relay
…PPP/Ethernet
Figures Courtesy of [5]
Label Creation
topology-based method—uses normal
processing of routing protocols (such as
OSPF and BGP)
request-based method—uses processing
of request-based control traffic (such as
RSVP)
Note:
OSPF-Open shortest-path first
BGP- Border Gateway Protocol
Label Spaces
Labels used by an LSR for FEC-label
bindings are split into 2 categories:


Per platform-label values are unique across
an entire LSR.
Per interface-label values are associated w/
interfaces. Label values provided on different
interfaces could be the same.
Label Distribution
No single method of signaling required

Enhancements of existing routing protocols (to allow
piggybacking of label information) include:
Border Gateway Protocol (BGP)
Resource Reservation Protocol (RSVP)

LDP (Label Distribution Protocol)- Defined by IETF for
signaling and management of label space.
--Extensions have been defined to support explicit
routing based on QoS and CoS requirements.
Label Distribution schemes
LDP—maps unicast IP destinations into
labels
RSVP, CR–LDP—used for traffic
engineering and resource reservation
BGP—external labels (VPN)
Signaling Mechanisms (general)
1.
2.
label request—An LSR requests a label from its downstream
neighbor so that it can bind to a specific FEC. This mechanism can
be employed down the chain of LSRs up until the egress LER (i.e.,
the point at which the packet exits the MPLS domain).
label mapping—In response to a label request, a downstream
LSR will send a label to the upstream initiator using the label
mapping mechanism.
Courtesy of [5]
Distribution and Signaling Protocols
•Implicit routing- labels are set-up and torn-down (like
telephone calls), also known as hard state.
•Explicit routing- allows for better traffic engineering, traffic
tunnels are created based on overall view of topology. More
dynamic.
Protocol
Courtesy of [10]
Routing
Traffic engineering
LDP
Implicit
NO
BGP
Implicit
NO
IS-IS
Implicit
NO
CR-LPD
Explicit
YES
RSVP-TE
Explicit
YES
OSPF-TE
Explicit
YES
Label Distribution Protocol (LDP)
Four message classes
1. Discovery-Announce and
maintain presence of an
LSR.
2. Session-establish, maintain,
terminate sessions b/w LDP
peers.
3. Advertisement-create,
change, delete label
mappings.
4. Notification-advisory and
error info.
•Discovery: Runs over UDP
•All others run over TCP
[9]
Message Structure
All LDP messages have a common message
structure (Type-Length-Value encoding scheme)
•Type: Type of message
[11]
For specifics on this frame see
http://www.networksorcery.com/enp/protocol/LabelDistributionProtocol.ht
m#Glossary
Discovery
LSR multicasts HELLO message to well-known UDP
port on “all routers on this subnet” multicast group.
All routers listen to this group to learn all LSRs with
direct connection.
When an LSR is detected, a TCP LDP connection is
established.
The HELLO message can also be sent to a wellknown UDP port at the IP address of a router if the IP
address is known through static configuration.
Some Important Messages
INITIALIZATION- label allocation mode, timer values, range of
labels to be used
KEEPALIVE- respond to Initialization of parameters are
acceptable. Connection is terminated if timely keepalives are not
received
LABEL MAPPING – Advertise a binding between address prefix
and label
LABEL WITHDRAWEL – reverse LABEL MAPPING, can occur
because of routing changes
LABEL RELEASE– Used in Conservative Label Retention mode
LABEL REQUEST– Used for down-stream-on-demand mode to
request label mapping
LABEL REQUEST ABORT – If next hop changes so that the prior
label request is invalid, this cancels the previous request
Slide courtesy of [9]
Performance Measurements
•LERs and LSRs:Juniper Networks M40TM
routers (MPLS and RSVP-TE).
• Interconnect:OC-12 (ATM 80 Mbps ATM
(PVC) connections for experiment.)
•Physical distance:
• LSR 1-LSR 3, LSR 2 -LSR 3 =40km
•LSR 1-LSR 2 =5km
Courtesy of [8]
•Computers: Pentium II 300
128 MB RAM
Fast Ethernet
FreeBSD 4.1
Results
Path from A to C (TCP stream)
Throughput of both MPLS paths
(TCP streams)
Courtesy of [8]
Results
Latency from A to C for TCP stream
Latency from A to C for UDP stream
Courtesy of [8]
Summary
Improves packet-forwarding performance in the network



MPLS enhances and simplifies packet forwarding through routers using Layer-2 switching
paradigms.
MPLS is simple, which allows for easy implementation.
MPLS increases network performance because it enables routing by switching at wireline
speeds.
Supports QoS and CoS for service differentiation


MPLS uses traffic-engineered path setup and helps achieve service-level guarantees.
MPLS incorporates provisions for constraint-based and explicit path setup.
Supports network scalability

MPLS can be used to avoid the N2 overlay problem associated with meshed IP–ATM
networks.
Integrates IP and ATM in the network


MPLS provides a bridge between access IP and core ATM.
MPLS can reuse existing router/ATM switch hardware, effectively joining the two disparate
networks.
Builds interoperable networks



MPLS is a standards-based solution that achieves synergy between IP and ATM networks.
MPLS facilitates IP–over-synchronous optical network (SONET) integration in optical
switching.
MPLS helps build scalable VPNs with traffic-engineering capability.
…However
Some Internet Purists complain that MPLS
breaks some critical Internet architectural
principles:


MPLS supports tunneling, which breaks the
transparency paradigm.
MPLS supports sessions, it breaks the
datagram model.
But MPLS provides great value to ISPs,
such as lower operating costs and ability
to provide QoS to businesses.
Questions???
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
Yin, Li, PowerPoint Presentation: “MPLS and GMPLS,” University of California,
Berkeley, Summer 2002.
R. Ramaswami and K. N. Sivarajan, Optical Networks: A Practical Perspective 2nd
Ed., Morgan Kaufmann Publishers.
Nortel Networks, “MPLS—An introduction to multiprotocol label switching,” 2001,
http://www.nortelnetworks.com/corporate/technology/mpls/collateral/55053.25-0401.pdf.
Semeria, Chuck, Juniper Networks, “Multiprotocol Label Switching: Enhancing
Routing in the New Public Network,” 2000.
International Engineering Consortium, “Multiprotocol Label Switching (MPLS),”
2003, http://www.iec.org/online/tutorials/mpls/
Farkas, K. et al. “IP Traffic Engineering of OMP Technique,” Technical University of
Budapest, Hungary, 2000.
Johnson, J., “Despite criticism, MPLS is here to stay,” Network World, April 2002.
http://www.nwfusion.com/columnists/2002/0408eye.html
Bayle, T. et al. “Performance Measurements of MPLS Traffic Engineering and
QoS,” Hiroshima University,
http://www.isoc.org/isoc/conferences/inet/01/CD_proceedings/T43/ .
Nortel Networks, “MPLS Tutorial,” May, 1999, http://www.nanog.org/mtg9905/ppt/mpls/ .
Gallaher, R, “Advanced MPLS Signaling,” December 2001,
http://www.convergedigest.com/tutorials/mpls3/page1.htm .
Network Sorcery Inc., “LDP,”
http://www.networksorcery.com/enp/protocol/LabelDistributionProtocol.htm#Glossa
ry .
Lines studied in
simulation
Simulation (EXTRA)
Courtesy of [6]
Results (EXTRA)
MPLS Simulation
OSPF Simulation
Courtesy of [6]
Results (EXTRA)
These simulations were done using an OMP (Optimized Multipath) extension
to their existing protocols.
OSPF-OMP
MPLS-OMP
Courtesy of [6]
History (Extra)
Control-driven model benefits




Labels are assigned and distributed before arrival of data traffic.
This means that if a route exists in the IP forwarding table, a label
has already been allocated for the route, so traffic arriving at a
multilayer switch can be label swapped immediately.
Scalability is significantly better than in the data-driven model.
Number of label switched paths proportional to the number of
entries in the IP forwarding table, not to the number of individual
traffic flows. Label assignment based on prefixes, rather than
individual flows, permits a single label to represent a highly
aggregated FEC.
In a stable topology, the label assignment and distribution
overhead is lower than in the data-driven model because labelswitched paths are established only after a topology change or
the arrival of control traffic, not with the arrival of each “new”
traffic flow.
Every packet in a flow is label switched, not just the tail-end of the
flow as in the data-driven model.