Side_1_360

Learning from other networks a series of studies
are documented in [ID_tems], although several
of the results are seen for ISDN-like traffic
behaviour. Some of the main observations are:

• Network performance seems to always be
  improved when TE methods are applied, and
  commonly a substantial improvement is seen.


• Sparse-topology multilink routing networks
  provide better overall performance under
  overload than meshed topology networks,
  although performance under failure may
  favour meshed topology with more routing
  alternatives.


• Using state information as in SDR provides
  essentially equivalent performance as using
  EDR. EDR is seen as an important class of TE
  algorithms, being adaptive and distributed in
  nature. Moreover, EDR may allow less over-
  head in terms of exchanging routing informa-
  tion.


• Bandwidth reservation is critical to stable and
  efficient performance of TE methods, and to
  ensure proper operation of multiservice band-
  width allocation, protection, and priority treat-
  ment. Bandwidth allocation per logical net-
  work (Virtual Network) is essentially equiva-
  lent to per-flow bandwidth allocation when
  looking at network performance and effi-
  ciency and allows great reduction in routing
  table management and size. This is also pro-
  posed as a trend in [Ov_NGI], ref. Figure 11.


• Single-area flat topologies give better network
  performance and design efficiencies compared
  with multi-area hierarchical topologies.


• Resource management is shown to be effec-
  tive to achieve service differentiation. MPLS
  bandwidth management and DiffServ queue-
  ing priority management are important for
  ensuring that performance objectives are met
  under a range of network conditions.


• Dynamic transport routing network design
  improves network performance in comparison
  with fixed transport routing for all cases
  examined in [ID_tems] for normal load pat-
  terns, abnormal load patterns and failure situa-
  tions.

Work on traffic engineering is going on in sev-
eral projects and standardisation groups. For
example, ITU-T formulated one question in
study group 2 addressing this topic. Seven
draft recommendations are under preparation:

• E.TE1: Framework for Traffic Engineering
  and QoS Methods for IP-, ATM- and TDM-
  based Multiservice Networks.


• E.TE2: Traffic Engineering and QoS Methods
  
  
  • Call Routing and Connection Routing Meth-
    ods.
  
  
  
  


• E.TE3: Traffic Engineering and QoS Methods
  
  
  • QoS Resource Management Methods.
  
  
  
  


• E.TE4: Traffic Engineering and QoS Methods
  
  
  • Routing Table Management Methods and
    Requirements.
  
  
  
  


• E.TE5: Traffic Engineering and QoS Methods
  
  
  • Transport Routing Methods.
  
  
  
  


• E.TE6: Traffic Engineering and QoS Methods
  
  
  • Capacity Management Methods.
  
  
  
  


• E.TE7: Traffic Engineering and QoS Methods
  
  
  • Traffic Engineering Operational Require-
    ments.
  
  
  
  

Some overall observations/results are captured

in Table 1.

6 Network Design Algorithm

The objectives of the algorithm for network

dimensioning are to obtain the capacities of

routers and link sets in this network and to find

the number and paths of LSPs that give the low-

est total network cost, including control, switch-

ing and transmission costs. Introducing LSP

capability, the question of whether or not to

cross-connect LSPs arise. Cross-connecting

traffic flows is a means of separating the traffic

flows from their previous LSPs that terminate in

the router and putting them into a new LSP that

could be cross-connected in this router.

6.1 Initial Steps

To investigate for a better LSP network, trade-

offs between control and switching costs and

costs for having separate LSPs should be bal-

anced. Typically, additional costs by introducing

more LSPs come from dividing the link set

capacities into smaller units (optionally reducing

the scale effect). Compared with some other

approaches (e.g. [E.737], [COST242], [Røhn97],

[Popp00]), the algorithm applied does not use a

global optimisation formulation. Rather, the pro-

cedure has some resemblance with the decompo-

sition approach in the sense that decisions with

respect to traffic handling and capacities are

made for each location in sequence, iteratively.

A flow chart of the initial steps is depicted in

Figure 12.