Side_1_360

• Priority Queuing within CBWFQ: The priority
  queuing within the CBWFQ feature brings the
  strict priority queuing functionality of IP RTP
  Priority required for delay-sensitive, real-time
  traffic, such as voice, to CBWFQ.

A schematic illustration of an outgoing interface
in a router is depicted in Figure 10.

The different serving policies have their charac-
teristics and corresponding preferred area of
applicability. Normally, Head-Of-Line (HOL) is
used such that real time traffic is given priority
over elastic traffic. The problem is that this high-
priority traffic would cause starvation for lower
classes during high load. On the other hand,
General Processor Sharing (GPS) disciplines
(such as WFQ) are preferred if a minimum band-
width must be guaranteed for each class. How-
ever, this kind of scheduling discipline is more
complex to implement than HOL, and is suscep-
tible to priority inversionif there is a higher-
class congestion. That is, a lower class can actu-
ally get a better service if there is congestion in
a higher class.

So, an observation may be that HOL is preferred
if higher priority classes demand is much lower
than lower classes demand. On the other hand,
GPS may be preferred in some cases if higher
priority classes demand is much higher than
lower classes demand. Besides, admission con-
trol can be introduced to limit the load in each
class to allow for bounds on the service levels.

When looking at some actual router implementa-
tions, one may find that several queueing and
scheduling steps may be arranged in series. For
example, on the output link, there may first be

queues per service class on the IP level. Then,

there may be queueing for placing packet flows

into a transmission system, and, lastly, queueing

for being transmitted on the link. For the last

queue a single First-In-First-Out queue is often

seen. Depending on the sizes and capacity of

service, all these queues may impact the actual

traffic flow characteristics, experienced delays,

effective service differentiation, etc.

2.4 Admission Control

Quoted from [COST257] admission control is

a preventive traffic control which aims to admit

an arriving new traffic source if and only if its

quality of service as well as that of the already

accepted sources is guaranteed. The admission

control procedure should also ensure a high

utilisation of network resources through efficient

statistical multiplexing.

Here a source may generate a set of flows.

Remembering that flowmay be defined as a uni-

directional succession of packets related to a

certain (part of an) application. Packets belong-

ing to the same flow have the same identifier

(e.g. given by source and destination addresses

and port numbers) and are initiated within a

maximum separation in time between each

other.

As stated, when running an application a number

of flows might result. An example is a multime-

dia application covering voice, video, file trans-

fers, interaction control, etc. All these flows

should be served in order for the application to

be run in a satisfactory manner. Hence, the term

session is introduced. A sessionis a continuous

period of activity during which a user generates

a set of flows (elastic or streaming type).

AF1 in of profile

AF1 out of profile

AF2 in of profile

AF2 out of profile

AF3 in of profile

AF3 out of profile

AF4 in of profile

AF4 out of profile

Best-Effort

HOL / GPS

LINK

GPS / HOL

EF (Premium)

Selective Discard

Out of profile packets have

a higher discard probabiliy

Figure 10 Example of out-

going link side in a router

when DiffServ is implemented