Telektronikk 2/3.2001
tination address with the corresponding part of
the address (longest prefix match) of every
neighbouring router.
When a new router is installed and the source
and destination of its in- and outgoing links are
defined, this information is passed to all other
routers by means of flooding. It is stored in the
topology database of the router and used to
recalculate the routing table entries affected by
the introduction of the new router. In this way,
each router holds information about the exis-
tence and topology of every other reachable
router, i.e. it has a view of the overall network.
Within the domain controlled by a single ISP,
i.e. within an Autonomous System, an Interior
Gateway Protocol like the Open Shortest Path
First – or the Intermediate System-Intermediate
System protocol is utilized. For interdomain
routing, a Border Gateway Protocol (BGP), such
as BGP-4 is used.
The forwarding function is realized as a FIFO
queue with tail-first dropping, i.e. in case of
congestion, the last incoming packet is dropped
first. There is only one queue for each output
port and no marking of packets, so every packet
gets the same treatment. Only one network ser-
vice is provided, Best Effort, i.e. “as soon as
possible” with “as much bandwidth as possible”.
There is no way of preventing greedy users
ignoring congestion signals and making things
worse for other users.
Most IP networks are Best Effort-based, despite
a number of shortcomings encountered with
such networks, in particular:
- There is no way of enforcing differentiated
behaviour to individual packets, flows or
traffic aggregates.
- There is no way of ensuring “fair” service pro-
visioning, i.e. binding the effect of over-usage
to the over-user.
- There is no way of providing differentiated
services with quantifiable QoS.
- There is no way of utilizing other routes than
the one chosen by the routing protocol, nor is
it possible to provide load-sharing with
slightly less favourable routes.
2.2 Traffic Engineering (TE) in IP
Networks
Traffic Engineering, in this context, relates to
the tools available to overcome the shortcomings
of Best Effort networks as previously described.
There are basically two options available for
achieving that, in terms of two network service
architectures:
- Integrated Services (IntServ);
- Differentiated Services (DiffServ), possible
in combination with MultiProtocol Label
Switching (MPLS) , i.e. DiffServ-over-MPLS.
Over-provisioning of network capacity may be
utilized temporarily to cope with expected traffic
increase, but it cannot solve any of the four
shortcomings of the Best Effort service on a
permanent basis.
An extensive discussion of the properties of
IntServ, DiffServ and MPLS is given in
[RessHandlIP].
2.2.1 Differentiated Services (DiffServ)
In DiffServ, aggregated packet flows are classi-
fied into Behaviour Aggregate (BA) traffic
classes. The treatment to be applied to all the
flows of a BA in every DiffServ router is called
the Per Hop Behaviour (PHB). The classification
may either be based on BA membership only, or
on Multiple Fields (MF) membership. With the
latter, a number of fields in the IP header, such
as destination host address and port number, are
used as additional basis for the classification.
The BAs are distinguished from one another by
the DiffServ Code Point (DSCP) values speci-
fied in the DS-byte of the IP4 header, replacing
the TOS byte when DiffServ is being used.
When multiple BAs share an ordering constraint,
i.e. the packets may not be re-ordered, these BAs
form an Ordered Aggregate (OA). The corre-
sponding set of PHBs is called a PHB Schedul-
ing Class (PSC). Behaviour may also be defined
on the AS level, so-called Per Domain Be-
haviour.
DiffServ presumes the existence of proper
SLAs/SLSs to define the service level to be
delivered to the users. The BA-/MF-classifica-
tion is part of the SLS.
A number of PHBs is being defined by the
IETF; Default Class (DC), Class Selector (CS),
Expedited Forwarding (EF), Assured Forward-
ing1-4 (AF1-4).
EF provides forwarding characterized by low
loss, low delay, low jitter and assured band-
width.
AF aims at delivering packets within the agreed
customer profile with high probability, whereas
out-of-profile packets are delivered whenever
available bandwidth permits. For each of the
four AF classes, three levels of drop precedence
Ståle Wolland (55) is Research
Fellow at Telenor R&D. He holds
a PhD in Physics from Heriot-
Watt University in 1974. He
joined Telenor R&D in 1976 and
has worked with information sys-
tems, distributed systems, net-
work management systems,
middleware and peer-to-peer
systems.
[email protected]
214
