Side_1_360

(Dana P.) #1
Telektronikk 2/3.2001

1 Introduction


From the outset, one may say the Internet Proto-
col (IP) was intended to transport information
from a source to a destination. The main empha-
sis may be that the information eventually
arrived with less strict requirements on time.
This may work well for some applications, like
computers and sensors exchanging data, and
when no impatient human being is involved.
However, as more types of applications are
loaded on the IP-based networks, additional
requirements are also placed on these networks.
Hence, the question arose how to efficiently sup-
port these applications. This became even
stronger as the commercial concerns grew for
providing services on the IP-based networks.

This may be the main motivation for proposing
the mechanisms described in this paper. Moving
beyond the best effort, other service models were
introduced; the Integrated Services (IntServ) and
the Differentiated Services (DiffServ) models.
Protocols for reserving resources were also pro-
posed, including the Resource reSerVation Pro-
tocol (RSVP). Avoiding the routing processing
and introducing means for load distribution and
traffic flow protection, the MultiProtocol Label
Switching (MPLS) was described. These are
described in the following chapters. Some basic
packet handling mechanisms are outlined first
(Chapter 2).

Quite a lot of variations and detailed information
is available on these subjects, and references are
pointing at the sources. In addition to the huge
background material refinements are steadily
on-going.

2 Congestion and


Packet Handling


Congestionis said to arise when too many pack-
ets (too high a load) are present in a sub-network
compared to the capacity of that sub-network.
Congestion has a tendency to feed upon itself,
for example as packets queued up may be timed
out and retransmitted adding to the congestion.
Another phenomenon is the spreading as un-
acknowledged packets will be buffered and
do not release memory space.

In this chapter means for managing congestion
are described. Basically, they are addressing
how packets are handled in an end-system and
in a router. The flow/congestion control mecha-
nisms in TCP are essential, as described in
[Jens01].

Congestion avoidanceis simply to avoid the
occurrence of congestion. Congestion avoidance
constitutes preventive congestion management
policies that can be categorised as having a long,
medium or short response time scale. Long-term
policies include capacity planning to expand net-
work capacity using estimates or forecasts of
future traffic demand and distribution. Medium-
term policies cover the minutes to days time
scale. Examples are adjusting routing parameters
and reconfiguring the logical network topology.
Short-term congestion avoidance includes packet
level processing functions as returned to later in
this chapter.

Congestion controlis commonly used to make
sure that a sub-network is able to carry the
offered traffic efficiently, involving all traffic
flows passing. On the other hand, flow controlis
used between a pair of sender and receiver pre-
venting the sender from transmitting packets too
fast, involving feedback from the receiver to the
sender. However, such a feedback may also be
set by the network, e.g. to avoid congestion,
implying that flow control mechanisms can be
utilised related to congestion control.

Several mechanisms (called policies in [Tane96])
affect congestion and how it could be dealt with,
see Figure 1, including:


  • At the end-to-end level: retransmission policy,
    out-of-order buffering policy, acknowledge-
    ment policy, flow control policy, timeout
    policy.

  • At the network level: use of connections, pol-
    icy for queueing and serving packets, discard
    policy (buffer is full), routing policy, packet
    lifetime policy.


Basic IP-related Mechanisms


TERJE JENSEN

When reading about IP-based networks, one almost immediately bangs into a series of abbreviations
and expressions. The main objective of this paper is to outline the basic mechanisms; Best Effort, Differ-
entiated Services, Integrated Services, MultiProtocol Label Switching, Resource reSerVation Protocol,
Label Distribution Protocol, and some of the ways packets are buffered and scheduled in a router.

Terje Jensen (39) is Research
Manager at Telenor R&D,
Kjeller. He earned his PhD
degree in 1995 from the Norwe-
gian University of Science and
Technology. Activities include
performance modelling and
analysis, dimensioning and net-
work evolution studies.
[email protected]

54
Free download pdf