Side_1_360

• At the link level: retransmission policy, out-
  of-order buffering policy, acknowledgement
  policy, flow control policy.

As seen from the list, policies for how the infor-
mation is carried in an accurate manner affect
the congestion management, as should be ex-
pected. In particular, appropriate means should
be taken to avoid that an emerging congestion
situation is further worsened, like simply to post-
pone the re-transmission of packets when the
buffers are filled.

A basic cause of congestion is that the current
traffic load is greater than the service capacity.
Commonly, this would be due to failures in the
service capacity (e.g. link failure) or that the
traffic is in a burst. The latter can be managed
by shaping the traffic, meaning that traffic peaks
are made smaller. This might be observed at the
egress (or supported at the ingress border), in
fulfilment with the agreed traffic pattern. At the
egress side this may be called a traffic enforcer.
On the other side of the border, the traffic flow
is frequently monitored and steps taken in case
the expected behaviour is not followed. Then a
traffic policercan be implemented, ensuring that
the resulting load is as expected. Leaky buckets
are frequently used for shaping and policing.
When variable packet lengths are used, like for
IP, the bucket may count in data volume, e.g.
bytes, rather than in number of packets. The
leaky bucket may operate on the data flow itself
or on tokens (being permissions to send data).
Suitable reactions must also be specified when
too much traffic is arriving. One reaction is to
drop the packet, another reaction is to change
the class/priority of the packet.

In order to set up a shaper and policer a proper
traffic flow specification has to be made. Such
a specification may use parameters describing
the traffic flow itself (mean bitrate, peak bitrate,
burst duration, etc.) or parameters more related
to a leaky bucket implementation (bucket size,

leak rate; see following section). When such a

specification has been done it may also be used

for admission control, i.e. to decide whether or

not more traffic should be admitted to the net-

work.

Buffering schemes and queueing disciplines are

important components in congestion handling. In

particular, when a number of service classes is

defined, those mechanisms have to maintain the

differentiation between the classes, e.g. by serv-

ing orders, packet dropping levels, and so forth.

Differentiations can be implemented for all

mechanisms described in the following sections.

2.1 Policing

Policing is a general term used for the process

of preventing a traffic flow from grabbing more

resources than allowed.Actions to be taken on

non-conforming traffic (packets) include drop-

ping the packet or remarking the packet. Re-

marking can be utilised for putting the packet

into another class/queue or to increase its proba-

bility for being dropped later in the network.

A leaky bucketalgorithm is frequently used to

define the conformance of a traffic flow. The

Figure 1 Levels/scopes for

congestion control

mechanisms

Figure 2 Analogy to leaky bucket

enforcer ensurer/policer

end-to-end

network

link

node

interface

Dropping

level

Tagging

level

Leak rate

Filling rate