107
Introduction
In the traditional Internet the only offered service
is a Best Effort service. The rapid traffic growth
makes the risk of over-provisioning rather low,
as what is over-provisioned today will likely be
insufficient in the near future. A common opin-
ion today is that IP networks will evolve towards
a new advanced architecture supporting also
classical Telecom services like voice, and
enabling service differentiation providing differ-
ent levels of performances. This will create new
business opportunities for Telecom operators
and also accelerate the process of renewing the
network infrastructures. To be successful, how-
ever, this process requires a greater capability of
controlling network performances. As a conse-
quence, there is an increasing interest toward the
definition and the implementation of techniques
exploitable to meet the desired level of perfor-
mance under different operational conditions.
This new field of activity is usually referred to
as Traffic Engineering for IP networks.Traffic Engineering (TE) is [IETF-TE] a control
process or a set of control processes that acts on
different time scales with the purpose of perfor-
mance optimisation of operational networks. In
a longer timescale this implies methods for net-
work planning and dimensioning, while in a
shorter timescale it implies control aspects of
routing and resource allocation.TE is often treated as virtually synonymous with
Multi-Protocol Label Switching (MPLS), but
work is ongoing in IETF to provide the neces-
sary features in IP routing protocols for what is
known as QoS routing or more generally con-
straint based routing. On the other hand, MPLS
has DiffServ support and is a hot candidate for
doing TE in DiffServ networks.This paper focuses on TE aspects related to tech-
niques that can be used for reconfiguring the net-
work in real-time. A main objective is to be able
to provide a consistent set of guidelines for con-
figuring a Differentiated Services IP network
based on the available technology, so that differ-
ent levels of QoS and different levels of service
reliability can effectively be met with an effi-
cient utilisation of network resources.Differentiated Services
Within IETF two IP service architectures have
been defined for the purpose of supporting dif-
ferent service demands with regard to network
capabilities, Integrated Services (IntServ) and
Differentiated Services (DiffServ). The IntServ
architecture [RFC1633] relies on the existence
of flow specific states that give the possibility
to reserve resources end-to-end and in this way
realise services with guaranteed performance.
The maintenance of these states, however, puts a
heavy burden on the IP routers as the state space
increases very rapidly, and IntServ is therefore
not seen as a scalable solution for future IP net-
works. But it is still a candidate for the access
part of such networks while DiffServ is a major
candidate for the core network. This is because
DiffServ is believed to be a more scalable way
to achieve QoS in an IP network since it acts on
aggregated flows and minimises the need for
signalling.The DiffServ architecture is defined in
[RFC2475]. It uses a new implementation of the
IP version 4 Type of Service (ToS) header octet.
This field is now called the DiffServ (DS) field.
It has 8 bits, out of which 6 bits are available for
current use and two are reserved for future use.
The available 6 bits define the DiffServ Code
Point (DSCP) and identify a Per Hop Behaviour
(PHB). The PHB indicates the way packets shall
be handled in the routers and can be set and reset
in any DiffServ capable router (marking). Such
handling can be delay priorities and drop prece-
dences.Some PHBs have been standardised; DE (default
class [RFC2474]), CS (Class Selector [RFC2474]),
EF (Expedited Forwarding [RFC2598]) and AF
(Assured Forwarding [RFC2597]). Each AF
class uses three DSCP values for differentiating
packets with different drop precedences (colour-
ing). This is mainly intended to be used in con-
nection with a congestion avoidance mechanism
in the routers in that packets may be dropped
based on a given probability that depends on
the actual buffer filling and the packets’ colour
(algorithmic droppers, e.g. Random Early Detec-
tion).Achieving Service Differentiation
in a Differentiated Services Network
by Use of MPLS
INGE SVINNSET
Inge Svinnset (46) is Senior
Research Scientist at Telenor
R&D. His research interests
include teletraffic, network per-
formance, Quality of Service
and dimensioning. During the
last ten years he has been
involved in several European
research projects related to
ATM network performance and
dimensioning. He is currently
working with Quality of Service
and Traffic Engineering in IP
networks.
Telektronikk 2/3.2001