Side_1_360

equivalent rate. The latter being a measure try-

ing to capture the variability of the rate, possi-

bly including other aspects, like information

loss ratio. Aggregates may also refer to all

traffic flows related to the same session.

• Information conveyance and locating func-
  tionality. Where are the functions located and
  how is the information carried between the
  different functions? Functions could be placed
  in the terminals/hosts, in the edge routers, in a
  dedicated server, and so forth. RSVP is a pro-
  tocol promoted for carrying the information
  between the functions, although others have
  also been promoted.

As mentioned earlier, executing the admission

control algorithm would basically provide an

answer whether to accept or reject a request for

serving a traffic flow. In principle the answer

could also be to accept but only on certain con-

ditions, e.g. that some characteristics of the

existing flows have to be changed/renegotiated.

Then in order for the algorithm to arrive at that

decision, a number of inputs has to be available.

Hence, the algorithms may differ in terms of

which inputs that are needed/taken into account.

Furthermore, the algorithms may also differ in

terms of which answers are possible (only accept

or reject, or more subtle outputs). An illustration

of inputs and outputs is given in Figure 11.

As shown, some likely inputs are:

• Characteristics of the new flow (which trig-
  gers execution of the admission control algo-
  rithm). A number of parameters can be used to
  characterise the new traffic flow, like peak bit
  rate, mean bit rate, requirements on delay, jit-
  ter and loss ratio, burst size, and so forth.
  These could be declared by the source or, in

principle, derived from other identifiers like

combinations of addresses, port numbers,

interfaces, etc.

• Characteristics of existing flows. The existing
  flows may be characterised in a manner simi-
  lar to the measures used for the new flow. The
  measures may be declared by the sources or
  estimated by monitoring.


• Measure of current load pattern for the re-
  sources considered. The current load on the
  resource can be monitored in order to obtain
  a more accurate measure of the situation.
  Applying such an input is commonly referred
  to as having a measurement-based procedure.


• User policy matters. A user profile may be
  available, e.g. stating on what service levels
  and under what conditions a traffic flow for
  that user is to be accepted. Some criteria are
  time-of-day, IP address, port number, inter-
  face identity, current load situation and char-
  acteristics of the new traffic flow.


• Resource policy matters. Certain policies on
  how to utilise the resources may be applied,
  such as acceptable load levels, use of over-
  booking, mixture of traffic flow types, and
  so forth.

For the admission algorithm various scopes and

principles could be valid, such as:

• What time scale is considered: Is only the cur-
  rent situation taken into account or is a more
  future-looking approach followed? Is the deci-
  sion to be based also on historic/trend infor-
  mation? An example is that a traffic flow
  accompanied by low revenue could be re-
  jected even with sufficient capacity available
  if there is a high probability that a flow
  accompanied by higher revenue will have
  to be rejected later on.


• What level of “gambling” is used for guaran-
  teeing the service level? Rather loose thresh-
  olds have to be used if strict guarantees are
  given, while tighter thresholds (and even over-
  booking) could be used when a more “gam-
  bling”-like attitude is taken.

2.4.4 Implementing Admission Control

In this section, a few examples are described

of how admission control can be implemented.

Each of them may not be completely satisfactory

for all the actors involved (e.g. user and network

operator). Moreover, some of them could well

be combined, like the RSVP-based and the pol-

icy-based.

Figure 11 Input and output
candidates for the admission
control algorithm

admission control

algorithm

time scale:

• short term


• long term


• historic/trends
  guarantee level:


• strict guarantee


• “gambling”

characteristics

new flow

reject accept on

conditions

accept

characteristics

existing flows

load situation

measure user policy

matters

resource

policy matters