|
Multi Label Protocol Switching (MPLS)
Introduction
It is now something of a cliché to talk about the "explosive" or
"exponential" growth of the Internet but the fact remains that it
has achieved remarkable growth. This growth has created a host of
technical challenges and Multi label Protocol Switching (MPLS) is
one of the responses to these challenges. MPLS evolved from a set
of developments by different vendors like IBM, Toshiba and is now
being standardized by the IETF body. This article will first introduce
to the challenges posed by the growth of Internet, how MPLS tries
to address them, basic framework of MPLS and references to the technology.
Challenges posed by Internet growth
Scalability: The classical routers in the Internet used to
do routing in software by making the longest compare of the addresses.
The routing algorithms were typically hand coded and crafted for
efficiency. The performance of these routers was acceptable for
a small traffic but the growth in IP traffic has made this architecture
obsolete and there is a need to go for switching in the core networks.
IP over ATM networks: In the early 90's,
ATM technology was touted as the universal technology and ISPs converted
their core network to ATM as it was based on switching giving better
performance. The mapping of IP over ATM posed challenges and new
protocols like MPOA, LANE were proposed to overcome the issues.
But the overlay of IP over ATM posed significant scaling problems.
The scaling problem basically means that the complexity of the network
grows much faster than the number of nodes in the network and after
some size; it becomes impossible to manage such a network. Since
the ATM is virtual circuit based, for a network of n nodes, it would
require n*(n-1)/2 virtual circuits for full connectivity. This becomes
unmanageable after some size.
Providing new classes of service: Initially
Internet was used as a data network and used for applications like
telnet and file transfer. As Internet became the defacto network
protocol, new and newer applications have been built on it. Some
demand stricter Quality of Service (QoS) guarantees for delay sensitive
voice / video applications while some demand tunneling services
like Virtual Private Network (VPNs). Today's Internet cannot support
these new applications easily.
Traffic Engineering: The ISP profitability
is heavily determined how it is able to use the network infrastructure
efficiently and specifically the available bandwidth. ISP should
be able to construct routes in such a way to avoid causing some
parts of the network be overutilized (congested) while some are
underutilized. Today's IP routing protocols are destination based,
least cost routing and do not take into account available bandwidth
in individual links as a parameter to determine the routing. This
is affecting ISP flexibility to engineer routes based on different
metrics and bandwidth available.
What is MPLS?
MPLS is an IETF specified framework for efficient designing, forwarding,
routing and switching of traffic flows in a network.
MPLS performs following functionalities:
 |
 |
Provides a mechanism
to map IP addresses to simple, fixed length labels used by different
packet forwarding/switching technologies |
|
Interfaces
to existing routing protocols like BGP, IS-IS |
|
Is independent
of L2 protocols like Frame relay, ATM or IP and hence allows
existing network to be upgraded for MPLS without hardware change |
MPLS Speak
A Label is a short fixed length identifier that is used to
forward packets. A label-switching device in the network will usually
replace the label in the packet to a new value before forwarding
to the next hop. A label is usually of local significance. There
are different protocols being proposed for the management of labels.
Label Distribution Protocol (LDP) is an IETF proposed protocol
that deals with assignment, management and distribution of labels.
Other alternatives include enhancing existing routing protocols
like IS-IS or BGP to distribute the labels or use RVSP to distribute
the labels.
In MPLS, data transmission happens on what are
known as Label Switched Paths (LSPs). The LSPs are a sequence
of labels at each and every node from the source to the destination.
LSPs are established either beginning of transmission or at the
time of detection of a traffic flow.
Since the labels are small and operations associated
with the labels are well defined, it is possible to implement the
label switching in the network in hardware.
The devices that participate in MPLS can be
of two types: Label Edge Routers (LERs) and Label Switching
Routers (LSRs). An LER operates at the edge of the MPLS network
and participates in establishment of LSPs and performs ingress/egress
MPLS operations. In case of ingress, it adds the label and sends
out to the MPLS network and in egress, it removes the label and
does the final forwarding. In typical applications, this will be
an aggregator edge device that takes in different type of traffic
like Frames, ATM and Ethernet, convert them to IP and does additional
MPLS functionality.
LSRs are high-speed router devices in the core
of an MPLS network that do label swapping and participate in creation
of LSP's.
Labels
A label is carried or encapsulated in the layer 2 header in the
network. MPLS offers flexibility to allow popular layer 2 protocols
to carry traffic without any change. In Frame relay, the Data Link
Connection Identifiers (DLCI) is used as a label. In ATM, the VPI-VCI
is used as the label. In general, a shim layer is added between
the data link layer header and IP header to allow carrying of labels.
Figure 2 gives general format of the label:
How does MPLS try to solve Internet problems?
Scalability: LSRs don't need to do longest match of addresses
in software and they can use the short fixed length labels to do
forwarding. Since labels are simple format, this operation can be
done in hardware and it improves the performance.
IP over any network: Since MPLS is L2
agnostic, any existing network can be upgraded to support MPLS.
In case of ATM, there is no need to construct lots of virtual circuits
reducing the complexity of current networks.
Different services over Internet: MPLS
provides Constraint Routed Label Distribution Protocol (CR-LDP)
that takes into account different QoS parameters like latency, Committed
Data Rate (CDR), Committed Burst Size (CBS) to reserve the resources
along all the nodes in the route. Other approaches like RSVP-Traffic
Engineering (RSVP-TE) also provide similar capability. This will
make sure that all LSR's in the MPLS network are able to treat the
traffic flows differentially and provide QoS.
Similarly, tunneling is a feature that is available
in MPLS. This is implemented using the label-stacking feature available
in MPLS. This helps in providing VPN functionality easily over the
network.
Traffic Engineering: Any solution to
the traffic engineering problem should establish routes that are
optimal with respect to scalar metric and also take into account
the available bandwidth on individual links. MPLS constraint routing
protocols like CR-LDP and RSVP-TE provide such a capability to use
the bandwidth more effectively.
Status of MPLS technology
Network vendors like Cisco, Nortel, Alcatel are rolling out core
routers with MPLS functionality. New network startups like Amber,
which are in aggregation-router business, are also adding MPLS edge
functionality to their boxes. It is expected that MPLS core networks
will be deployed more actively in 2001.
Caveats with MPLS technology
Though MPLS is touted as panacea to all
of today's Internet problems, it has couple of caveats:
|
|
For the benefits
of the MPLS to be realized, the core network routers have to
be upgraded to MPLS. Even if one of the devices in core network
is not MPLS, it may not work. |
|
MPLS does not
provide end-to-end IP QoS still. It addresses QoS in core networks,
but still gaps are there when IP packets travel from core to
access network to the end user. |
References
Rajagopal is the Director of the Network Edge
and Access Technology group at MindTree Consulting.
He can be contacted at n_rajagopal@mindtree.com
|
