20.11.2012 Views

Contents Telektronikk - Telenor

Contents Telektronikk - Telenor

Contents Telektronikk - Telenor

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

100<br />

change the routing for the next call.<br />

The DAR-pointer is not moved.<br />

- If the call attempt is blocked on the<br />

second leg of the DAR-alternative,<br />

(assuming E is the DAR-alternative,<br />

both EB and EC is blocked for DARtraffic),<br />

the call attempt will be lost. A<br />

signal is returned to A, stepping the<br />

pointer in the DAR list to the next<br />

alternative, F. This alternative will be<br />

the DAR-alternative for the next call<br />

A-X.<br />

DAR is a simple learning algorithm. It<br />

was first described by British Telecom.<br />

The principle is also used in a method to<br />

be implemented by NTT, Japan in 1994.<br />

Circuit reservation is mandatory for DAR<br />

to obtain network stability and protection<br />

of the direct routed traffic. In a simulation<br />

study carried out by Norwegian<br />

Telecom Research in 1988 [3] our top<br />

level mesh network was simulated with<br />

different routing strategies under different<br />

load conditions like normal traffic<br />

load, skew load, focused overload and<br />

general overload. The main conclusions<br />

were:<br />

- The method gives a flexible network<br />

with good traffic flow also when there<br />

is considerable difference between<br />

traffic actual offered and dimensioning<br />

traffic/network resources. The network<br />

will be robust towards traffic variations<br />

and wrong dimensioning.<br />

- The method distributes the congestion<br />

justly.<br />

- The network throughput will not<br />

degenerate, but is robust towards general<br />

overload. In this case the routing<br />

converges towards direct routing.<br />

- The method achieves most of the possible<br />

benefits. Compared to more complex<br />

adaptive methods, there are no<br />

significant differences.<br />

- The method is simple and does not<br />

require major data collection from<br />

other parts of the network. The additional<br />

processor load will be small.<br />

- The dimensioning of the network will<br />

not be complicated. The network may<br />

be dimensioned as for direct routing.<br />

In [4] G.R. Ash from AT&T presents a<br />

comparison between 4 routing strategies<br />

- Learning with random routing (DAR)<br />

- Real time state dependent routing<br />

(RSDR – least loaded via route selection<br />

based on real-time network status)<br />

- Periodic state dependent routing (least<br />

loaded via route selection based on<br />

periodic network status)<br />

- Sequential routing with flow optimisation<br />

(time-variable engineered routes<br />

plus overflow routes updated in realtime<br />

(Example DNHR – Dynamic<br />

Non-hierarchical Routing).<br />

A large (103 node) and a small (10 node)<br />

mesh network were simulated with high<br />

day load, global overload, focused overload<br />

and cable failure. The basic conclusions<br />

were that<br />

- real-time state dependent routing<br />

(RSDR) significantly outperforms the<br />

other strategies in large networks<br />

- DAR and RSDR performs comparably<br />

in small networks<br />

- all dynamic routing strategies outperform<br />

current hierarchical routing by a<br />

large margin<br />

- the cost/complexity of DAR may be<br />

lower than the other strategies.<br />

As the Norwegian network is a small network<br />

these conclusions from 1994 are<br />

consistent with our simulation study in<br />

1988.<br />

9 Dimensioning criterion<br />

<strong>Telenor</strong> Nett distinguishes between<br />

- Normal load dimensioning utilising<br />

criterion for normal traffic load:<br />

⋅ Congestion dimensioning, maximum<br />

congestion at normal traffic load<br />

(An ), e.g. 0.5 or 1 %<br />

⋅ Standard dimensioning, in addition<br />

to criteria for congestion dimensioning,<br />

a criterion for maximum utilisation<br />

(erlang/circuit), U, has to be fulfilled<br />

(normally U = 0.8). The objective<br />

is to make large circuit groups<br />

more robust towards overload.<br />

- Fault tolerant dimensioning (or reliability<br />

dimensioning) utilises criteria<br />

both for normal load and failure situation.<br />

For failure situations the circuit<br />

group is dimensioned to fulfil a<br />

defined congestion level at dimensioning<br />

traffic during failure, Af .<br />

Af = An * Co * Cr where<br />

Co = Overload factor<br />

= traffic offered the circuit group<br />

during failure / normal traffic<br />

load on the same circuit group<br />

Generally Co = 2.0 for dual homing,<br />

1.5 for triple homing. Co has to be estimated<br />

individually for each circuit<br />

group in the top level mesh network<br />

Cr = Traffic reduction factor<br />

= Dimensioning traffic during failure<br />

/ Normal traffic load<br />

Some measurements have indicated that<br />

in 94 % of the total number of hours in a<br />

year, the traffic offered will be less than<br />

80 % of the ITU-T normal traffic load.<br />

We have decided to define the network<br />

as unavailable when the call blocking is<br />

higher than 20 %. As a first attempt our<br />

dimensioning criteria were set to ensure<br />

that the call blocking in the trunk network<br />

should be less than 20 % (B =<br />

20 %) during a failure situation if the<br />

traffic offered does not exceed 80 % (Cr = 0.8) of the normal traffic load. The<br />

modular dimensioning of circuit groups<br />

in addition to yearly network expansion,<br />

will, however, give some extra capacity<br />

to most circuit groups.<br />

When determining criteria for fault tolerant<br />

dimensioning and in which parts of<br />

the network the method should be used,<br />

the models in ITU-T Rec. E.862, Dependability<br />

planning of telecommunications<br />

networks, have been very useful.<br />

The use of high usage circuit groups will<br />

be rather limited, mainly restricted to circuit<br />

groups between large EOs in the<br />

same LT-area with high mutual traffic<br />

interest.<br />

There are no plans to change the dimensioning<br />

criteria when circuit reservation<br />

is introduced as long as only a small proportion<br />

of the traffic is given priority.<br />

10 Optimal combination of<br />

robustness methods<br />

In the conventional PSTN network few<br />

systematic methods have been used to<br />

obtain a certain level of availability.<br />

Automatic restoration in the transmission<br />

level is, however, common for the main<br />

long distance systems. Also alternative<br />

routing in PSTN and occasional use of<br />

diversity routing increase the availability.<br />

In the target network the logical and<br />

physical (SDH – Synchronous Digital<br />

Hierarchy) network structures fit well<br />

together. An example of a SDH-ring in<br />

an LT-region is shown in Figure 10. In<br />

the example the exchanges (EOs and<br />

LTs) are connected to the transmission<br />

ring by Add and Drop Multiplexors<br />

(ADM). Systematic use of double hom-

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!