Side_1_360

tive combinations can be examined. It means

that several approximations could be introduced

and compared.

6.2 Dependability Concerns

Strategies for establishing protection/backup

paths for parts of LSPs and utilisation of such

paths may have great implications on depend-

ability and also on the cost of the network de-

sign. As seen from an operator’s view point, a

solution with pre-allocated restoration paths and

sharing of restoration capacity between working

LSPs seem to be flexible and cost efficient.

Then, a set of attributes as described in

[Awdu99] can be attached to each LSP, like

resilience attribute, pre-emption attribute, adap-

tive attribute, etc. The backup LSP should be

router disjoint from the working LSP and can

be established without reserving capacity. This

is done as the backup LSP may be common to

more than one LSP and these LSPs are likely to

have different bandwidth requirements, and the

backup LSP pool may be common to many

backup LSPs.

In the following it is assumed that two depend-

ability traffic flow classes are used; high and low

priority. In case of failure the low priority class

may be dropped from the links carrying the

backup LSP in case the bandwidth is not suffi-

cient to carry the total load. Each link must then

have access to a backup capacity that is at least

the maximum of the needed capacity for high

priority flows on the working LSPs that will use

the link on its backup LSP. A backup capacity

may be the difference between the total capacity

and the capacity needed to carry the link’s high

priority traffic flows.

The part of the algorithm related to dependabil-

ity starts after the initial steps described in the

section above. First, all links are considered one

by one, and for each link (“failed” link) all LSPs

are examined. A backup LSP is identified for the

high priority traffic flows having the same end

points as the considered LSP on the link. The

LSPs on the link are examined according to

decreasing cost. As a mixture of low and high

priority traffic may flow on an LSP, the needed

capacity to restore the high priority traffic is cal-

culated first. Then a route for a backup LSP is

found by applying a shortest path algorithm

(Dijkstra’s algorithm). The “distance” measure

used in that algorithm is then the cost of trans-

mission and switching. In case some links hav-

ing available restoration capacity is looked at the

corresponding restoration capacity is attached

with zero cost. However, when doing these cal-

culations all needed capacity for high priority

traffic on the “failed” link must be taken into

account. Finding a backup LSP is only done

once for an LSP, meaning that for some LSPs on

a link the above calculations may already have

been done when that link is examined.

7 Examples

The numerical results given in this section are

based on an example network depicted in Figure

13. The network consists of routers that are
    placed at 14 locations in Norway.

The network structure is described by the fol-

lowing parameters:

• Location identity – given by the name of the
  city where the router is located;


• Relative demand – presented by a percentage
  of the total traffic generated;


• Unit cost per link between two locations.

Five applications are considered and charac-

terised in Table 2. For simplicity all applications

are assumed to use a call handler, implying that

blocking requirements could be more relevant.

The total demand per application is also given

Figure 13 Network structure,
giving location identity,
relative demand and unit cost
per link between locations

Trondheim

9

Lillehammer

Ålesund^9

4

Bergen

10

Stavanger

9

Kristiansand

6

Arendal

7

Tønsberg

6 Sarpsborg

5

Gjøvik

5

Bodø

5

Tromsø

5

Oslo 1

(^14) Oslo 2
6
287
378
170
245
69
452
837
427
478
243
168
80
53
91
320
15
124 167
44
(^3421217)
1274
554
554
Servers for the Telegame
application in Arendal, Oslo 1,
Bergen and Trondheim