A Proposal for the Specification of a SVC Service in a MPLS UNI
EDSON MOREIRA SILVA NETO, SERGIO VIANNA FIALHO
Graduate Program in Electrical Engineering
Federal University of Rio Grande do Norte
Centro de Convivência – Campus Universitário, Lagoa Nova – Natal – RN
BRAZIL
Abstract: - New multimedia applications that use the Internet as a communication media are pressing for the
development of new technologies, such as: MPLS (Multiprotocol Label Switching) and DiffServ, that
introduce new and powerful features to the Internet backbone, as the provision of QoS (Quality of Service)
capabilities. However, to obtain a true end-to-end QoS, it is not enough to implement such technologies in the
network core, it becomes indispensable to extend such improvements to the access networks, what is the aim
of works presently under development. To contribute with this process, a new and powerful service schema is
presented herein as a mean to support a true end-to-end QoS, through the extension of MPLS. Thus, it is
specified a Switched Virtual Connection (SVC) service to be used in the context of a MPLS User-to-Network
Interface (MPLS UNI), that is able to efficiently establish and activate Label Switched Paths (LSP), starting
from the access routers, that satisfy the QoS requirements demanded by the applications. The benefits and
most important issues to be considered when using this specification are also included.
Key-Words: - Access Network, MPLS, SVC, UNI, QoS
1 Introduction
When IP (Internet Protocol) was defined for the
first time, the current networks requirements were
barely imaginable. Nowadays, besides supporting an
exponentially increasing volume of data, carried by
the traditional Internet services as file transfer and
electronic mail, networks should be able to
differentiate among several classes of traffic, to
support the needs of multimedia applications related
to the transport of voice, music and video, under
demand or even in real time. These demands have
led to a search for QoS (Quality of Service) in the
Internet, and became an essential requirement for
the success of the trend of global convergence in the
use of IP as the communication media, either in the
core of long distance networks or in the access
networks.
Due to the requirements introduced by these new
applications, many changes are to be expected with
respect to the model proposed for the Internet. The
traditional hop-by-hop IP routing schema, for
example, is beginning to reach its technological
limit, what implies a necessary paradigm change in
the packet forwarding process, once even the high
performance routers recently produced have proven
to be insufficient to deal with the current traffic in
the Internet core.
So, label switching technologies techniques
appear as efficient solutions, to increase the speed of
packet forwarding, providing additional capabilities
to differentiate between classes of traffic and
support to QoS. MPLS (Multiprocol Label
Switching) [1], in particular, is a standard defined by
IETF (Internet Engineering Task Force) for label
switching and allows, among other functionalities,
the creation of Label Switched Paths (LSP), with
different service characteristics. Such capability
improves significantly the process of packet
forwarding, due to the simplification of the IP
routing process, as long as it avoids the need to
repeat packet header processing in every hop of the
path. MPLS also provides for QoS support.
So, there is strong evidence that future multiservices networks shall operate over IP/MPLS
technologies, offering new opportunities for TE
(Traffic Engineering) and VPN (Virtual Private
Network). It should still be taken into account, that
all this services improvement is expected to be cost
effective and is likely to create new business
opportunities for ISPs (Internet Service Provider).
There should be no doubt that MPLS provides a
solid base for the new demands with respect to the
use of network resources in the core, providing
excellent QoS mechanisms. However, a new
bottleneck should be found in the access networks.
Therefore, this works proposes to investigate
how to extend the benefits of MPLS to the access
network, more specifically to the AR (Access
Router), by means of the specification of a SVC
(Switched Virtual Connection) service for a MPLS
UNI. A PVC (Permanent Virtual Connection)
service for a MPLS UNI is already defined [2]. The
specification proposed herein intends to endow the
AR with the capability to request LSPs with specific
QoS parameters for the sake of establishing
switched connections.
Figure 1 presents the scope of the proposed
service.
PVC
LAN
AR
ISP
LER
SVC
Access
Network
UNI MPLS
MPLS
Domain
Fig. 1 – Scope of the SVC Service
2 SVC Service Specification
The definition of a MPLS UNI provides an
interface for connection between a CPE (CustomerPremisses Equipment), actually an Access Router,
and a public MPLS network. This interface
guarantees that a MPLS-based infrastructure can
support critical performance applications, including
real time exchange of voice and video preserving the
use and flexibility of the current Internet
technologies.
For such a task, the MPLS/FR ALLIANCE has
already sponsored a work group, responsible for the
development of the referred UNI, with the objective
to provide true end-to-end QoS over MPLS-based
networks. A MPLS UNI will also allow the
operation of traffic management tools within
customer premises, for they were restricted to the
backbone core. The above mentioned group has
already developed the first version of the MPLS
UNI [2], which contemplates only the PVC service,
where the initiative to establish a new connection is
taken by the service provider, reserving the AR a
passive operation. However, the SVC service is still
not specified. To establish or close switched
connection, ARs should be capable to send service
requisitions involving LSP’s management.
It remains a challenge to provide good levels of
QoS for the access networks, where there is less
control and reliability, and where it is expected the
heavier constraints on VoIP (Voice over IP) and
videoconference services.
2.1 Initial Considerations
The SVC service proposed in this work should
include, at least, the following characteristics:
scalability, simplicity and functionality.
Scalability. It is important to bring to the AR the
advantages due to the granularity with which FECs
(Forwarding Equivalence Class) can be defined
within MPLS. This allows for traffic aggregation
and guarantees a individual end-to-end QoS without
the need to maintain status information related to the
flow in all network segments of the core, thus
alleviating the processing load within LERs (Label
Edge Router).
Simplicity. It is desirable to restore the Internet
initial paradigm, known as the end-to-end principle,
meaning that complexity should be left for the final
systems while keeping the network simple.
Functionality. It is also of interest to facilitate
LSP establishment procedures, initiated from the
access network, with the addition of certain
attributes as, for example: destination address and
QoS constraints.
It is still possible to list two aspects related to the
SVC service specification that need to be analyzed
in further detail: the choice of the signaling protocol
to be used by the service and the choice of the
bandwidth allocation mechanism for each switched
virtual connection.
a) Choice of the signaling protocol.
This should consider label distribution issues, as
well as the needs for QoS provision, implying the
reservation of network resources in an appropriate
way along a LSP. For this purpose, it would be
possible to choose among several options, including:
LDP (Label Distribution Protocol) [3] and the
RSVP-TE (Resource Reservation Protocol – Traffic
Engineering) [4]. The protocol recommended in this
work is the RSVP-TE, due to the following
characteristics:
Security. It implements authentication and
explicit policy control. The PATH and RESV
messages carry an object that contains the desired
policy in a protected way: this object’s content can
be used to perform admission control based on
politics;
Soft-state Protocol. It automatically registers
routing changes and, implements an efficient
mechanism against losses of control messages. The
refreshing mechanism helps to assure that the state
of LSP is properly synchronized among neighboring
hops; in case of changes, for example, in rerouting
or as the demand for network resources within a
LSP is modified (a change in traffic parameters),
RSVP answers faster, since all hops related to the
LSP maintains state information;
Creation of ER-LSPs (Explicity Route) that
provides soft rerouting, preemption and loop
detection features.
b) Choice of the bandwidth allocation mechanism.
Until lately, IETF TE-WG has defined two
models of bandwidth restriction: Maximum
Allocation Model (MAM) [05] and Russian Dolls
Model (RDM) [06]; besides those, we are
considering the use of Artificial Intelligence
techniques, for example a neural network to capture
the dynamics of the traffic profile in the network
and, based on this knowledge, deciding, in an
automated way, on the allocation of network
resources.
service – that is to say, they should agree on a traffic
descriptor, or on characterization of the traffic to be
transported. This can be interesting from the point of
view of ISPs, because it opens new business
possibilities for them and consequently assures a
greater use of the installed network infrastructure.
Further studies are required concerning billing
issues on the use of this service. From the customer's
point of view, it is also interesting to count on SVC
services, because one can apply for additional
services, whenever necessary.
Once the on-line contract has been “signed”, it is
understood that the network agrees in supporting the
traffic with the specified characteristics and also that
the subscriber agrees in not exceeding the
performance limits allocated to him.
Figure 2 shows the distribution of the functions
related to QoS provision among the several network
elements.
2.2 Suggested Functionalities
SVC service should be capable to provide, at
least, the following functions:
a) The subscriber should be able to initiate the
establishment of new LSPs, specify related QoS
requirements, as the rate of packets loss and jitter,
and also specify traffic parameters as: medium
throughput, peak throughput, peak duration, and
others;
b) Once an on-line contract has been signed,
presuming, that both parts have agreed upon the
terms of SLA (Service Level Agreement), the LSPs
should be activated;
c) Both the subscriber and the network provider
should be able to close active LSPs;
d) Support modifications on QoS parameters, that is,
support signaling procedures that allow the AR to
modify LSP’s attributes, since the SLA is not
disregarded;
e) Support layer two encapsulation especially for the
Ethernet technology. New studies can include
encapsulation of another protocols.
2.3 Overview of SVC Service
A crucial point in this proposal is related to the
definition of a strategy for traffic control: one should
determine if a new LSP can be accommodated
within the network, in accordance to the
performance parameters requested by the customer.
Once the network guarantees the allocation of
sufficient resources to attend the customer's
solicitation, the subscriber and the network provider
should establish an on-line contract concerned the
temporary traffic on the LSP reserved for the SVC
Queuing
Scheduling
Congestion Control
Access
Network
LSR
AR
Classification
Marking
Policing
Shaping
LER
Policing
Queuing
Scheduling
Core
MPLS
Fig. 2 – QoS Provision Functions
a) Functions implemented by AR (Access Router):
Classification. Packets are classified in the AR,
according to QoS requirements. Thus, the
processing overhead time due to classification
purpose is decreased in LERs. Additionally, labeled
incoming packets will release ARs from
classification procedures and diminish the
processing time spent with Layer-3 forwarding
tables lookup.
Marking. The bits of the MPLS packets header’s
EXP field are marked in accordance with the CoS
(Class of Service) chosen during classification.
Naturally, together with the classification, marking
procedures should also be implemented in the AR
incoming interface.
Policing. The rate of packets arrival should be
metered for each class, in order to guarantee that it is
in agreement with SLA conditions, avoiding a
contract break and consequent penalties. Then, the
objective of this procedure is to police the execution
of that agreement and also to avoid congestion.
Shaping. It involves the adaptation of the actual
traffic to the agreed profile. That is particularly
important, because, once a LER and/or LSRs along
the path can aggregate traffic, the shaping provided
by the AR will guarantee that different classes of
traffic won't interfere with each other. This aspect of
the traffic management procedures is based on an
efficient
token-bucket
and
leaky-bucket
implementation, as described in [7].
b) Functions implemented by LER (Label Edge
Router):
Policing. Policing is important in LER,
considering the ISPs’ need to guarantee that the
traffic originating from a subscriber is conforming
with the SLA. It also provides data accounting
purposes. Admission control procedures are also
necessary; if one expects to be able to reject new
demands for LSPs when the available resources risk
to be exceed.
Queuing. Use of different queues to
accommodate the traffic from several service
classes.
Scheduling. Procedures adopted for servicing of
to those queues.
c) Functions implemented by LSR (Label Switched
Router):
Queuing, Scheduling and Congestion Control.
Inside the network core, packets are queued in
agreement with the EXP field. Possible scheduling
techniques include: WFQ (Weighted Fair Queue),
RR (Round Robin), PQ (Priority Queuing), among
others. Thus, each egress interface determines,
based on CoS requirements, when a packet should
be discarded. The most used algorithms are: RED
(Random Early Detection) and WRED (Weighted
Random Early Detection). Yet, it is worth noticing
that, in the core, policing is dispensable, because one
can presuppose that the traffic comes from reliable
sources.
2.4 “Modus Operandi”
Once the label distribution protocol is
operational, the AR can, whenever necessary,
request labels for LSPs that are likely to be
provisioned in the egress interface. Moreover, once
LSPs are unidirecional, labels need to be allocated
for each direction.
The AR initiates the signaling process by sending
a PATH message to the LER, specifying the desired
LSP attributes. The addressed LER, after verifying if
it can provide for the present solicitation, answers
with a RESV message, also informing the label to be
used in the AR-to-LER direction. This procedure
should then be repeated in the opposite direction
(from-LER-to-AR). The LSPs are distinct but they
are correlated by the association of their identifiers,
resulting in a pair of bi-directional correlated LSPs.
Figure 3 illustrates the basic sequence of messages
exchange for the SVC service establishment.
1st – Path (LER)
AR
LER
2nd – RESV – Label x
3rd – Path (AR)
AR
LER
4rd – RESV – Label y
Fig. 3 –Message exchange for the SVC service
establishment
In a SVC service, AR and LER act
simultaneously as RSVP peers, changing the
message categories specified by RSVP-TE.
With relation to the label distribution and
management procedures, SVC service should use
the conservative retention mode [1], once it is
necessary to maintain only the connections that have
been selected by the AR; and, to use the DOD
(Downstream On Demand) method when triggering
the distribution of labels, that is to meaning that
labels are requested by the AR only when necessary.
3 Foreseen Benefits
Some of the more evident benefits associated to
the use of the proposed service were already
identified and are described in the following:
a) Higher performance for applications running
directly over MPLS. Since a LSP has already been
established for a given application, its data flow is
transported directly over MPLS by redirecting it to a
specific port. Transporting application data (such as
VoIP) directly over MPLS would eliminate the
overhead time due to the use of other forwarding
protocols, resulting in better performance levels with
respect to latency and reliability.
b) QoS Management per Service. This characteristic
facilitates the implementation of true end-to-end
QoS, because it defines explicit mechanisms for
resource management and network control
according to previously agreed parameters, as long
as it is possible to make reservation for different
LSPs for each service class. MPLS also helps in all
traffic management procedures (metering, policing,
shaping and marking) at LSP level.
c) Homogeneous Control Plane. A homogeneous
control plane throughout the network is the better
way to manage it. Being capable of using the same
set of tools throughout the whole network minimizes
problems of interoperating between different types
of equipments. From ISPs’ point of view
homogeneous transport mechanisms are seen as a
more efficient and flexible way to cope with the
network cost structures: both costs with the network
infrastructure and operational costs. Besides, it
allows the datacom provider services portfolio to be
expanded [8]. This will enable the datacom
companies to dynamically adjust its network
demands based on factors such as: the traffic load
per application (bandwidth allocation) and the
application performance (QoS/CoS parameters).
d) Flexibility in the establishment of new billing
systems. MPLS gives flexibility to telecom and
datacom concerning the implementation of new
accounting schemes, based on the real behavior of
subscriber’s applications.
4 Conclusion
Considering that a demand for QoS provision is
irreversible, in IP networks, with a strong tendency
for the use of RSVP as a resource reservation
protocol and DiffServ mechanisms for traffic
classification and, considering still that the core
networks are migrating to a IP/MPLS solution, one
can conclude that access networks are likely to
become the new bottleneck in data transport with
respect to the provisioning of true end-to-end QoS.
So, it seems to be highly relevant the proposition of
a SVC service, specified in the context of a MPLS
UNI. Such a service constitutes a new, powerful and
flexible means of extending MPLS from LERs, at
the border of ISPs’ networks to the customers’
access routers assuring a series of additional benefits
for both ISPs and subscribers, as for example: higher
performance, QoS management per service and
homogeneous control plane.
The main techniques envisaged for a reliable
SVC service were discussed and some critical issues
were addressed, in order to contribute to a thorough
description of the service.
References:
[1] E. ROSEN, et al, Multiprotocol Label Switching
Architecture, Internet RFC 3031, 2001.
[2] MPLS PVC User to Network Interface –
Implementation Agreement. MPLS/Frame Relay
Alliance. MPLS/Frame Relay Alliance Technical
Committee. Maio, 2003. Extracted from
www.mplsforum.org.
[3] L. ANDERSON, et al, LDP Specification,
Internet RFC 3036, 2001.
[4] D. AWDUCHE, et al, RSVP-TE: Extensions to
RSVP for LSP Tunnels, Internet RFC 3209,
Dezembro, 2001.
[5] F. Le Faucher, Maximum Allocation Bandwidth
Constraints Model for Diff-Serv-aware MPLS
Traffic Engineering, draft-ietf-tewg-diff-te-mam00.txt, Jun. 2003.
[6] F. Le Faucher, Russian Dolls Bandwidth
Constraints Model for Diff-Serv-aware MPLS
Traffic Engineering, draft-ietf-tewg-diff-terussian-03.txt, Jun. 2003.
[7] MPLS e next generation access network, White
Paper. Integral Access. Extracted from
www.integralaccess.com.
[8] Layer 3 switching Using MPLS, White Paper.
Extracted from www.netplane.com.