Side_1_360

analyses varying the weight factors. In principle,

these weight factors can be considered as taking

part of the cost or placing relative credibility on

each, e.g. when certain components are esti-

mated with higher accuracy.

5 Some Lessons Learned

Elsewhere

While IP-based networks do have their charac-

teristics, looking at other networks may well

give several ideas on how to efficiently config-

ure and manage the network. In most traditional

telephone networks, a fixed hierarchical routing

has been applied. A number of results have indi-

cated that introducing more dynamic schemes

may improve the blocking and the network

resilience. Three main types of dynamic routing

have been described:

• Time-dependent routing (TDR); changing
  effective routing tables at pre-planned time
  instants, e.g. due to daily variations in the
  traffic pattern;


• State-dependent routing (SDR); changing
  routing tables according to the network state,
  given e.g. by traffic load;


• Event-dependent routing (EDR); changing
  routing tables triggered by certain events, like
  congestion thresholds crossed.

All these types can be introduced in the routing

policies, although the routing protocols would

likely take care of the situations that may arise.

In one respect, if arrival of a routing message is

called an event, the routing could be categorised

as event-dependent when looking at an individ-

ual router.

Figure 11 A common
infrastructure carrying a
number of logical networks,
adapted from [Ov_NGI]

Circuit-switched (-like) traffic tunnel – carries

existing circuit-switched telco traffic

High priority network transit traffic – uses IP

QoS to enforce performance guarantees

Low priority transit traffic – best effort general

IP traffic (e-mail, Web, ftp, etc.)

High priority customer traffic – uses Diffserv to

select service levels

Includes circuit-switched voice and

virtual circuit traffic (e.g. LAN

interconnect) that cannot be handled

by a general IP service for technical

or security reasons

Includes inter-network traffic with

agreed QoS between operators,

wholesale IP services

Includes general subscription or free

traffic delivered as a best effort

service (e.g. most web browsing)

Includes priority traffic delivered to

premium customers or users

designated by content providers as

“preferred users”

Assuming there are Npaths that could be used

for establishing the connection, where path khas

blocking probability Pbk, this cost component

could be written as:

where connections belonging to aggregate q

have a mean offered traffic of Aqwith mean

duration hq. The cost factors δ, γand εexpress

the effort related to a successful connection, an

additional search for paths and rejection of a

connection, respectively. As recognised, this

expression is adapted from circuit switched net-

works. Depending on the router mechanisms

available, more than one path may not be select-

able. Then the expression becomes fairly simple.

Total costis obtained by adding the different

contributions after assigning different weights.

In this way the cost for a group of flows, Q, can

be calculated as:

where kt, ksand kcare relative weight factors

related to transmission, switching and control

cost. The relative weighting of the cost compo-

nents is a difficult issue that may have signifi-

cant implications on the logical network design.

It is therefore essential to perform sensitivity

ZQ= {}ktZtq+ksZsq+kcZcq

q∈Q

∑

Zc Zc

A

h

Pb N Pb

iPbPb

qb

bq

b

b

k

k

N

bq

k

k

i

i

i

N

=

= ()− ++()

⎡

⎣

⎢

⎤

⎦

⎥

⎧

⎨

⎪

⎩⎪

++() ()−

⎡

⎣

⎢

⎤

⎦

⎥

⎫

⎬

⎪

⎭⎪

∈

∈ =

=

+

=

−

∑

∑ ∏

∑ ∏

δγε

δγ

1

1

1

1

1

1

1

1