Computer Simulation of Computer Networks
CE321
Network Engineering
“beyond Cisco CCNA” material Part 2
v 2.2
CE321 – beyond Cisco notes ‹#›
1
Part 2: Multiprotocol Label Switching (MPLS)
CE321 – beyond Cisco notes ‹#›
2
IP uses hop-by-hop forwarding paradigm
IP is connectionless
Routers forward packets hop-by-hop using destination address according to routing table in each router
Source and destination addresses in packet header are unchanged by routers
Packet is of variable length
CE321 – beyond Cisco notes ‹#›
3
IP routing table example
Routers perform longest prefix match to determine next hop for packet
Packet destined for 197.8.3.54 matches both routers T and Y in the Z’s routing table
Y is chosen as it has the longest match
Routing table at Z
Network Gateway
197.8.0.0/22 T
197.8.3.0/24 Y
197.8.2.0/24 T
CE321 – beyond Cisco notes ‹#›
4
IP routing protocols
Routing table in each IP router updated by routing protocols:
Interior gateway protocols
Disseminate topology or routing information among the routers
From this, each router determines its routing table
Open Shortest Path First (OSPF) or Routing Information Protocol (RIP)
Used within an Autonomous System (AS)
Exterior gateway protocols
Allow an AS to advertise its reachable networks to another AS
Topology of AS not advertised
Border Gateway Protocol (BGP)
Used between ASs
IP routing protocols periodically update the routing tables
Truly distributed route determination algorithms
Nodes calculate next hop for packets, based on information from neighbours
CE321 – beyond Cisco notes ‹#›
5
Routing loops and failures
Distributed algorithm does not necessarily provide up-to-date information
Routing errors may occur
Can result in packets looping endlessly
Avoided in IP by routers decrementing the time to live field in the IP header
When it reaches zero, packet is discarded
When a failure is discovered, the algorithm converges to suit the new network topology
As the algorithm is truly distributed it is resilient to node failure
This is part of the driving ethos behind the evolution of IP networks
CE321 – beyond Cisco notes ‹#›
6
ATM networks
Different information types require different qualities of services from network
In terms of bandwidth, delay, loss
Real-time traffic
Video, telephony
Non real-time traffic
Email, file transfer
Telephone networks support a single quality of service
Expensive to implement and run
In its original form, Internet supports no quality of service
Flexible and cheap
ATM networks were meant to support a range of service qualities at a reasonable cost
ATM was connection-oriented
ATM transported fixed-size packets (cells) over virtual circuits
CE321 – beyond Cisco notes ‹#›
7
Label switching: ATM
ATM is connection oriented, providing virtual circuits
Header defines virtual path (VPI) and virtual channel (VCI)
Labels, swapped at each switch according to switching table
Not necessary to carry entire destination address
Cells are of fixed length (53 bytes)
CE321 – beyond Cisco notes ‹#›
8
Worked Example – labels
ATM has a 12-bit label field for implementing “virtual” paths
How many possible labels are there?
Suppose that labels cannot be re-used, and that each bi-directional circuit in the network must have a unique label assigned to it along its entire length
In a complete mesh of paths among N nodes, what is the maximum possible number of nodes?
Now suppose that label swapping takes place at each node, and that a label can hence be re-used in different parts of the network
Assume that the minimum capacity of a circuit (in each direction) is 30 Kbyte/s and that the maximum capacity of a link in each direction is 622 Mbit/s
How many labels would be required to ensure unique identification of each path?
CE321 – beyond Cisco notes ‹#›
9
Answers
212 = 4096 possible labels
N(N – 1)/2 ≤ 4096 so solve N(N – 1) = 8192
Hence N2 – N – 8192 = 0
So N = (1 + √(1 + 32768))/2 = 91.011
Therefore the maximum number of nodes is 91
Maximum number of paths on link =
622 × 106/(30000 × 8) = 2591.67
Therefore a maximum of 2591 labels will be required
CE321 – beyond Cisco notes ‹#›
10
Virtual circuits
Label saves on header space
Needs to be pre-established
Needs to be swapped at intermediate points
Need translation table and connection setup
All packets must follow the same path
Switches store per-VCI state
Can store QoS information
Separation of forwarding and control (route determination)
Virtual circuits do not automatically guarantee reliability
Small IDs can be looked up quickly in hardware
Harder to do this with IP addresses
CE321 – beyond Cisco notes ‹#›
11
Label swapping vs hop-by-hop forwarding
Label swapping needs simple one-to-one mapping
Input label/port to output label/port
Use content addressable memory
IP forwarding uses longest prefix match
Requires search through routing table
May be of the order of 100,000 entries
More complex algorithm
More software intensive
Switches generally have a better price/performance ratio
To achieve fast line rates with IP forwarding requires very fast processors and/or parallel architectures
Growth of IP requires increase in line rate over time
Would rather use switching than IP forwarding if only considering price/performance of transport components
One solution runs IP over a switched network using an overlay
CE321 – beyond Cisco notes ‹#›
12
IP overlay network
IP routers with ATM cards are connected at the IP layer through ATM virtual channels
Switching is performed in the high speed core
IP forwarding is performed at the slower network edge
CE321 – beyond Cisco notes ‹#›
13
Problems with overlay network 1
Requires management of both an IP and ATM network
IP and ATM have evolved as isolated systems
Integration of IP and ATM is problematic
There are many different systems for achieving IP over ATM (e.g. LANE, CLIP, MPOA, …)
Although IP over ATM gives more cost effective switching it has lower bandwidth utilisation
IP must be encapsulated in ATM cells
Many IP over ATM solutions require complex servers to achieve integration
Gives a single point of failure and can limit scalability
CE321 – beyond Cisco notes ‹#›
14
Problems with overlay network: example
Ingress and egress are through the same ATM switch
However the IP datagrams have to pass through the IP forwarding of router B
IP and ATM do not use information from each other
IGP thinks shortest route is through B
Some IP over ATM techniques supply a cut-through mechanism to get around this problem
These have increased complexity
CE321 – beyond Cisco notes ‹#›
15
Problems with overlay network 2
Another problem is the number of router adjacencies
Consider N routers
Require full mesh to reduce bottlenecks
Single router hops between every ingress/egress
Each router has N – 1 adjacencies
N(N – 1) ≈ N2 (for large N) virtual circuits required
If there is a failure, or change in the network, routing algorithms send information to other nodes to converge
Each router sends an LSA for all the other N – 1 routers
Each other routers re-floods this to N – 2 neighbours
Update information can hence grow as a factor of N4
Generic problem with, for example, OSPF
This solution has scalability problems
CE321 – beyond Cisco notes ‹#›
16
Additional new requirements for IP routing
High speed and cost effective IP forwarding is required
However there are also other new requirements:
End-to-end quality of service guarantees
Moving away from the current best effort IP quality of service
Traffic engineering facilities
Such as load balancing across routers, rate limiting via policing, ….
Support for virtual private networks
Traditionally, IP route determination (or control) and forwarding components are closely integrated
Changing functionality of one always affects the other
Adding this new functionality requires both the forwarding and control components to be updated in all routers
These added facilities place extra burden on forwarding component
A considerably more complex algorithm is required than just longest prefix match
Want to address new requirements without complicating forwarding
CE321 – beyond Cisco notes ‹#›
17
Problem with tight IP forwarding/control integration: example
Traffic from A and B is destined for F
If forwarding decision is only based upon destination all traffic for F passes either all through D or all through E
Conventional destination based IP routing does not allow load balancing across D and E
More advanced forms of certain routing protocols and router implementations, e.g. OSPF, allow a limited form of load balancing
Not adequate for more complex traffic engineering
CE321 – beyond Cisco notes ‹#›
18
Worked Example – capacity allocation
All traffic in this (highly artificial) network is from A to G (500 Mb/s), A to H (40 Mb/s), B to G (400 Mb/s) and B to H (50 Mb/s)
Without MPLS, OSPF at router C sends all traffic destined for G via E, and all traffic for H via D
If all links have capacity 622 Mb/s, what happens?
With MPLS, each demand can be routed via either D or E
How many ways in total are there of routing all demands, assuming unlimited link capacity?
State at least one way of routing label switched paths which is compatible with the constraint of 622 Mb/s for each link
A
B
C
D
E
F
G
H
CE321 – beyond Cisco notes ‹#›
19
Answers
The traffic over C-E-F is 500 Mb/s + 400 Mb/s > 622 Mb/s
The links are overloaded and congestion occurs
The traffic over C-D-F is 40 Mb/s + 50 Mb/s << 622 Mb/s
The links are underutilised
There are 4 demands therefore 24 = 16 ways of routing all of them
If demands A-G and A-H are routed via D, 540 Mb/s is carried over C-D-F
Also, demands B-G and B-H are routed via E, so 450 Mb/s is carried over C-E-F
This avoids overloading or underutilisation of links
CE321 – beyond Cisco notes ‹#›
20
Problems to be solved
Need to evolve the routing architecture of IP networks
There is a need to add new routing functionality (QoS, load balancing etc)
IP is the dominant end-to-end protocol
It makes sense to optimise the network structure to support it
Need better performance or price/performance ratio in routers
Mapping IP onto ATM or Frame Relay is problematic
Overlay networks can introduce routing scaling problems (N4)
Need to have a highly scalable solution
Many see the price/performance issue and overlay N4 problem as mostly historical
Modern IP routing systems have greatly improved
IP is carried over a transport network (typically SDH/SONET) that is unaware of QoS, traffic engineering requirements or routing
Adding new services and simplifying management is now seen as the main driving force for change
The suggested solution is Multiprotocol Label Switching (MPLS)
The control function of IP routing is now separated from the forwarding function
CE321 – beyond Cisco notes ‹#›
21
Previously considered overlay architecture
This has problems as discussed before
CE321 – beyond Cisco notes ‹#›
22
New architecture with MPLS
MPLS nodes perform switching
Under control of normal IP routing protocols
CE321 – beyond Cisco notes ‹#›
23
History behind MPLS
A number of bodies realised the problems of IP over ATM
They all came up with comparable schemes:
Use ATM switching hardware as the core forwarding component
Discard ATM signalling
It is difficult to map IP control onto ATM signalling
Replace it with control protocols that map into the IP domain
The schemes include:
Toshiba's cell switching router – prototype demonstrated in 1995
Ipsilon's IP switching – set out as open standards
Cisco's tag switching – standards as part of the IETF
Supports switching over several link layer technologies (not just ATM)
IBM's aggregate route-based IP switching (ARIS) – similar to tag switching
These proprietary solutions started a new IETF working group that has developed into MPLS as we know it today
CE321 – beyond Cisco notes ‹#›
24
MPLS network architecture
CE321 – beyond Cisco notes ‹#›
25
Network components
Label switch router (LSR)
A node, forwarding labeled datagrams according to a table
Table maps incoming labels to a specific outgoing interface
Table defines the new label for the packet when it leaves the node
MPLS uses label swapping as in ATM
Edge label switch router (ELSR)
A node device, accepting incoming unlabelled packets
Places suitable labels onto the packets
So that they can be forwarded through the MPLS network
At the egress it strips off the label and forwards the packet using conventional IP routing
All packets that belong to a forwarding equivalence class (FEC) are mapped to the same label
CE321 – beyond Cisco notes ‹#›
26
Forwarding Equivalence Classes (FECs)
An FEC defines a set of packets that have attributes that can be uniquely determined
These attributes are common to all packets in the FEC
All packets that share the same destination address prefix could be part of the same FEC
A single application flow could form an FEC
Taken from source address, destination address and destination port
Could involve inspection of headers for UDP, TCP and/or RTP
Even application header fields in theory
Examples above demonstrate extremes of FEC granularity, choice is up to system designer
Using application flow could exhaust the label space in a core network
CE321 – beyond Cisco notes ‹#›
27
Worked Example – FECs
A network has N ELSRs, each with L local area networks attached to it via IP routers
Each LAN supports U users running A applications each
How many LSPs must exist in the network if they exist between:
All pairs of ELSRs?
All pairs of LANs?
All pairs of users?
All applications of the same type between each pair of users?
Pick representative values for N, L, U and A, and work out the number of LSPs in each case
Comment on the feasibility of finer FEC granularities
CE321 – beyond Cisco notes ‹#›
28
Answers
Choose N = 50, L = 4, U = 32 and A = 8
Mesh between ELSRs: N(N – 1)/2 = 1,225
Mesh between LANs: LN(LN – 1)/2 = 19,900
Mesh between users: LNU(LNU – 1)/2 = 20,476,800
Mesh between similar applications: LNUA(LNU – 1)/2 = 163,814,400
Finer FEC granularity results in a very large number of LSPs and may overload the label space
220 labels = 1,048,576
CE321 – beyond Cisco notes ‹#›
29
MPLS and the ISO OSI reference model
Mapping MPLS onto the ISO OSI protocol stack model is problematic
The OSI model does have limits of usefulness
Not layer 2 – independent of layer 2 technology
Not layer 3 – has no routing and addressing of its own
Not really a “layer” – no single format for transporting data from layer above
CE321 – beyond Cisco notes ‹#›
30
MPLS control architecture
MPLS control architecture defines how label switching forwarding tables are updated in each LSR
Binding represented by ordered pair: (label, FEC)
CE321 – beyond Cisco notes ‹#›
31
MPLS forwarding component
MPLS changes the conventional IP forwarding model
In IP the control component (e.g routing algorithm) and forwarding component are closely integrated
Leads to problems of network evolution (discussed before)
MPLS separates the forwarding from the control component
Label switching forwarding component uses a single simple forwarding algorithm based on label swapping
The label carried in a packet is short, has a fixed length, and has no structure; it uniquely specifies forwarding and resource attributes
Label switching forwarding component does not place any restrictions on the granularity associated with a label
Label switching forwarding component can support multiple network layer protocols as well as link layer protocols
All this has been abstract so far we can now ask: what is a label?
CE321 – beyond Cisco notes ‹#›
32
MPLS label
An MPLS label only has local significance
Significance only between two neighbouring LSRs
It can be anything that can be used in a label swapping switch architecture
Examples of systems that can support a label in existing link layer headers:
ATM VCI field
ATM VCI and VPI fields
Frame Relay Data Link Circuit Identifier (DLCI)
For other link layer protocols that cannot support labels in the link layer headers (e.g. Ethernet) MPLS defines a shim label
This is inserted between the link layer header and the IP packet:
Successive LSRs may use different label encodings
For example, LSR with both ATM and Ethernet interfaces
CE321 – beyond Cisco notes ‹#›
33
Label switch forwarding table
The label switch forwarding table maps incoming labelled packets by label value to the appropriate output interface and outgoing label
The table is effectively a pool of available labels, some of which are assigned to FECs and provide forwarding
Some labels may be unassigned, awaiting a new FEC definition that requires a label
The table is updated by the control components using a label distribution protocol (LDP)
Incoming label Outgoing label Next hop Outgoing interface
1 3 197.0.3.24 if0
2 4 192.2.6.32 if1
3 2367 192.0.3.24 if0
4 - - -
5 63 192.2.5.56 if2
etc etc etc etc
CE321 – beyond Cisco notes ‹#›
34
Label switch forwarding table example
Consider table entries on each router for FEC representing destination 192.6/16
Table entries use the default hop-by-hop IP route
Assume that conventional IP routing protocols have updated the routing table in each LSR
These reflect the network topology
Thus, each LSR has a routing table entry for the network 192.6/16 which is used by MPLS to identify a FEC
CE321 – beyond Cisco notes ‹#›
35
Label switch forwarding table example (continued)
Initially each LSR knows nothing of labels at other LSRs
Each LSR locally binds an available label to the FEC by putting it in its table
Outgoing label is not yet known
Router Incoming label Outgoing label Next hop Outgoing int
A 100 ? B if1
B 6 ? E if1
C 17 ? D if2
D 5 ? E if0
E 22 ? E if0
Each line represents one entry from each of the five label tables
CE321 – beyond Cisco notes ‹#›
36
Label switch forwarding table example (continued)
Each LSR distributes local bindings to adjacent LSRs
Ordered pair (FEC, label)
Consider A sending bindings to B and C
B and C know from IP routing tables that A is not the next hop to the destination so no table updates are made
Router Incoming label Outgoing label Next hop Outgoing int
A 100 ? B if1
B 6 ? E if1
C 17 ? D if2
D 5 ? E if0
E 22 ? E if0
Each line represents one entry from each of the five label tables
CE321 – beyond Cisco notes ‹#›
37
Label switch forwarding table example (continued)
Consider B sending bindings to A, D and E
For D and E, B is not the next hop to the destination so the information is ignored by these LSRs
But, B is the next hop for A so A takes this remote binding and adds it to its table (see below)
Router Incoming label Outgoing label Next hop Outgoing int
A 100 6 B if1
B 6 ? E if1
C 17 ? D if2
D 5 ? E if0
E 22 ? E if0
Each line represents one entry from each of the five label tables
CE321 – beyond Cisco notes ‹#›
38
Label switch forwarding table example (continued)
Distribution of label information continues between adjacent nodes
Final table entries shown below
Node E is an edge LSR and thus there is no outgoing label entry for this node
It strips off the label and uses conventional IP forwarding
Router Incoming label Outgoing label Next hop Outgoing int
A 100 6 B if1
B 6 22 E if1
C 17 5 D if2
D 5 22 E if0
E 22 ? E if0
Each line represents one entry from each of the five label tables
CE321 – beyond Cisco notes ‹#›
39
Label switch forwarding table example (continued)
Once the bindings are complete a label switched path (LSP) is said to be established
Packet transport may start using label swapping as shown
If the FEC is no longer required, label bindings are deleted
For example, routing change due to a fault
CE321 – beyond Cisco notes ‹#›
40
Note, it is possible for labels to be reused
Labels 100, 6 and 22 are rearranged on C-D-E
Label 6 is assigned to part of the path on D-E
This is possible if router B has a per-interface label space
No ambiguity between label 6 on if0 and on if2
Ambiguity results if router B has a per-platform label space
CE321 – beyond Cisco notes ‹#›
41
Label distribution protocols
MPLS requires a mechanism to transport the label binding information
A label distribution protocol (LDP) is required
MPLS may be used in various network scenarios and with different forwarding functionality requirements
Hence there is not one but many different LDPs specified
Default LDP is the new protocol defined in MPLS for disseminating labels for the conventional hop-by-hop destination based IP routing
Label distribution on BGP (BGP-LDP) “piggybacks” label bindings as an extension of the existing BGP protocol
An exterior gateway IP routing protocol
RSVP-LDP adds label bindings to the existing RSVP protocol, which is used for signalling QoS reservations in IP routers
Constraint routed LDP (CR-LDP) is a new protocol for disseminating label bindings in a network that requires QoS
Alternative to RSVP-LDP
Only LDP is considered in detail in these notes
CE321 – beyond Cisco notes ‹#›
42
The MPLS label distribution protocol
MPLS-LDP will be used to refer to the new default label distribution protocol defined within MPLS
Must be distinguished from the generic term LDP
In some texts the term LDP is used ambiguously
MPLS-LDP has several components
These implement LSR peer discovery
Configure communication between neighbours
Four classes of messages:
DISCOVERY messages – learn of existence of neighbours
ADJACENCY messages provide initialisation, keepalive and shutdown sessions between adjacent LSRs
LABEL ADVERTISEMENT messages are used to send label binding advertisements, requests, withdrawal and release
NOTIFICATION for advisory and error information
DISCOVERY messages transported over UDP
All other messages transported over reliable TCP
Like most IP-based protocols it is designed to be extensible
Through type, length, value (TLV) encoded objects
CE321 – beyond Cisco notes ‹#›
43
MPLS-LDP (continued)
DISCOVERY message:
HELLO message is sent to well-known UDP port on “all routers on this subnet” multicast group
All LSRs listen and thus learn about existence of neighbours
ADJACENCY messages:
INITIALIZATION messages allow LSRs to agree on and define:
unsolicited downstream vs downstream on demand label assignment
ordered vs independent LSP control
liberal vs conservative label retention
KEEPALIVE messages sent periodically to show that LSR is still active and that parameters are acceptable
LABEL ADVERTISEMENT messages:
LABEL MAPPING messages send information about a label binding
LABEL WITHDRAWAL messages remove a label binding
e.g when routing table entry removed due to change in route
LABEL REQUEST message is used in downstream on demand label assignment to demand a label binding
CE321 – beyond Cisco notes ‹#›
44
Concept of label stack
CE321 – beyond Cisco notes ‹#›
45
Concept of label stack (continued)
MPLS supports multiple labels organised as a stack
This allows the creation of a LSP tunnel
In previous slide there is one tunnel with two levels of hierarchy
E1, B1, B2 and E2 are peers on one level of the hierarchy
B1, X, Y, Z, B2 are peers on the second level of the hierarchy
B1 and B2 are gateways between the two routing hierarchies
B1 has a FEC to destination address beyond E2 and B2 supplies a label binding for this route (L3)
However, B2 is not the next hop in the internal network
Hence B1 must push L3 onto the label stack and request a new label binding for the FEC that represents the address of B2
A key applications of a LSP tunnel is for transit networks (and VPNs)
For example, OSPF sets up tunnels between BGP speakers
iBGP peers use the tunnels via label stacks
Internal routers need not be aware of external BGP routes
Significant performance advantages
Amount of IP forwarding is reduced
Separates routing inside and outside the AS
Operator can control routing independently
CE321 – beyond Cisco notes ‹#›
46
Worked Example – LSP tunnels
Two LSP tunnels share a number of consecutive links along their respective routes
They both start and end at different LSRs
Draw a diagram to illustrate this
What happens if two LSPs each pass through one of the tunnels using the same label?
If these links carry two tunnels, how many LSPs in total can exist on each, including those inside the tunnels?
CE321 – beyond Cisco notes ‹#›
47
Answer
Even although both LSPs use label 567, they are not confused along C-D-E since these routers only look at the outer label – one for tunnel A and one for tunnel B
All the other labels apart from those two can be used
220 – 2 = 1,048,574 in total
There are 220 possible LSPs within each tunnel, so total is 1,048,574 + 1,048,576 + 1,048,576 = 3,145,726 LSPs
tunnel A
tunnel B
LSP X
label 567
LSP Y
label 567
A
B
C
D
E
F
G
CE321 – beyond Cisco notes ‹#›
48
Providing customer connectivity through MPLS
MPLS is widely used in operator networks
Traffic can be broken into two categories:
Internet access for broadband and corporate customers
traffic between customer sites (site-to-site virtual private connections)
These notes will concentrate on the second category
Aside: in some ways the first case can be considered as a special case of the second where the operator is its own customer and the service is simply Internet access for all its customers.
CE321 – beyond Cisco notes ‹#›
Site-to-site connections using MPLS
Three (common) ways to provide connectivity between customer sites using MPLS
MPLS L3 VPN (BGP/MPLS VPNs RFC4364)
MPLS L2 virtual private LAN service (VPLS) using BGP (RFC4761)
MPLS L2 VPLS using LDP (RFC4762)
These notes will consider the first two (the third may be considered similar to the second – as far as a customer is concerned)
CE321 – beyond Cisco notes ‹#›
General model for site-to-site MPLS
CE – Customer Edge router (usually owned an managed by the customer)
PE – Provider Edge router, an edge label switch router
P – Provider router
A label stack two-deep is used at the PE router
First, inner, label is only used at the PE routers to identify customers
Second, top, label, is used to forward between PE routers via the P routers
IP/MPLS
PE
attachment circuit
customer 1 site
customer 1 site
customer 2 site
customer 2 site
Packet Switched Network (PSN) IP or MPLS
PE
CE
CE
CE
CE
Demarcation points
P
P
CE321 – beyond Cisco notes ‹#›
51
MPLS L3 VPN (BGP/MPLS VPNs RFC4364)
Providers manage the WAN routing and export routes to the customer using some means e.g. using an interior gateway protocol (IGP) between CE and PE routers, or through statically configured routes
If using an IGP, this may be OSPF, but an instance only used for this purpose, ie not the OSPF instances used for internal routing in either the customer or the provider networks. A customer’s main IGP in each site is kept separate (ie IGPs in different sites do not peer)
Provider uses BGP in the PE routers (but not the P routers)
Each customer has a different VPN Routing and Forwarding (VRF) table in the PE routers. BGP has the ability to give each VRF a different route distinguisher (RD), ie typically a unique RD for each customer. This allows different customers to use the same IP address ranges – as long as they do not need to communicate between themselves.
Provider manages the IP routing for the customer between the customer sites.
This means the customer does not have the effort of managing the WAN routing.
CE321 – beyond Cisco notes ‹#›
BGP as a label distribution protocol
PE routers form internal BGP (iBGP) peers in a full-mesh to share the VRFs
VRFs ensure customers CEs have the correct routes to access their other sites (but do not have the whole topology of the remote sites)
iBGP is used for PE label distribution (as well as sharing routing information)
BGP is a highly flexible protocol so the label information is shared in a BGP extension
iBGP peers are only PE routers (not P routers) so the BGP-based label distribution protocol only controls the distribution of the inner, customer, labels
LDP (or some other) is used to configure the top, outer, labels
Use of label stack (inner for customer, outer for core) has benefit:
PE routers have large routing tables (the many VRFs for each customer)
P routers only need to have an IGP routing table (e.g. using OSPF) for the core network and do not need to hold the VRF routes
CE321 – beyond Cisco notes ‹#›
MPLS L2 VPLS using BGP (RFC4761)
End-to-end architecture
Allows MPLS networks to provide multipoint Ethernet services
It is “Virtual” because multiple instances of this service share the same physical infrastructure
It is “Private” because each instance of the service is independent
Isolated from one another
It is “LAN Service” because it emulates Layer 2 multipoint connectivity between subscribers
Connections between PE routers are called pseudo-wires (PW)
Provider has no L3 routing communication with customer
Customer has to manage all the WAN routing between the sites
CE321 – beyond Cisco notes ‹#›
54
VC label identifies interface
Tunnel label(s) gets to PE router
Unidirectional Tunnel LSP between PE routers to transport PW PDU from PE to PE using tunnel label(s)
Both LSPs combined to form single bi-directional pseudo wire
Directed LDP session between PE routers to exchange VC information, such as VC label and control information
VC distribution mechanism using LDP
IP/MPLS
PE1
LSP created using IGP+LDP or RSVP-TE
customer site
customer site
customer site
customer site
Label Switched Path
iBGP between PE1 and PE2
PE2
CE
CE
CE
CE
CE321 – beyond Cisco notes ‹#›
55
A PWES is either: - an Ethernet link or a VLAN link between two ports, or - an ATM VC or VP, or - a Frame Relay VC, or - a TDM circuit, or - an MPLS LSP
Note that the PSN tunnel may be MPLS, L2TP, GRE and so on .. UTI is another mechanism to transport the PDUs between ingress and egress PE – in this case the PW is created using a UTI tunnel.
Typical VPLS frame
DA
SA
Type=
0x8847
PW
CW
Top
Label
FCS
Inner
Label
DA
SA
Type=
0x8100
VLAN
Type=
0x0800
IP Datagram
Ethernet
MPLS
Payload (an Ethernet frame)
One possible Ethernet frame
PWCW – Pseudo wire control word is 4 bytes (provides sequence number, a type field for some control frames and other features)
Note the two MPLS labels (as defined earlier)
Large overhead means Ethernet “Jumbo Frames” are used by provider (9000 byte payload.
CE321 – beyond Cisco notes ‹#›
Summary: MPLS L3VPN and VPLS (using BGP)
MPLS L3VPN VPLS (using BGP)
Customer does not manage WAN routing tables. IGPs at each customer site are completely separate. Customer has to manage all the WAN routing, the MPLS network appears as a single Ethernet bridge.
Customer obtains routing information from provider as static routes, or from an IGP between CE/PE. No routing protocol required between customer and provider
Said to scale well as IP routing can scale to a large number of sites Has some scalability problems if not designed well as broadcasts have to go to all customer sites
MPLS L3VPN and VPLS similarities
MPLS used in the core network with two layer stack (inner identifies customer)
BGP used to distribute customer labels in PE routers
IGP and LDP (or RSVP-TE) used to manage internal, provider, label distribution
Both likely to use Ethernet as access technology for customers
Differences
CE321 – beyond Cisco notes ‹#›
/docProps/thumbnail.jpeg