It is thus seen that the XFG algorithm would in
this case use three objective functions: an objec-
tive function based on the M/G/1/Kqueue to
compute the price of bandwidth for signalling;
the Erlang-B formula to compute the price of
bandwidth for conversational and streaming traf-
fic; and an objective function based on the TCP
fixed point of Section 5.1 to compute the price
of bandwidth for interactive and background
traffic. The bandwidth market would use the
three objective functions to find and capacitate
the LSPs based on the profits to be earned by
carrying the various service classes.
In addition it may e.g. be preferred to distinguish
network management traffic and, in a split archi-
tecture, traffic related to the communication
between servers (MSCs, TMSCs, SGSNs and
GGSNs) and gateways (MGWs). These two
classes would be handled by the same type of
objective function as signalling, but with differ-
ent performance requirements. Typically, the
network management information may be rela-
tively insensitive to delays, whereas server-gate-
way traffic may have strict real time require-
ments.
From a conceptual point of view, the LSPs can
be viewed as members of logically independent
networks of which there is one for each service
class. The logical networks may for example
include RNCs, MSCs and TMSCs for conversa-
tional services; RNCs, SGSNs and GGSNs for
interactive services; and RNCs, MSCs, TMCSs,
SGSNs, GGSNs, HLRs, EIRs etc. for signalling
traffic.
The latter observation indicates that the method
can be extended to include VPNs in a straight-
forward manner. The procedure is to characterise
the requirements on VPN links by objective
functions and represent each VPN link by an
aggregate after which XFG will find suitable
routes and bandwidths. Also note that the
method can be applied on top of existing VPNs.
7 Conclusions
MPLS represents a promising way of obtaining
the benefits of different technologies such as
ATM (with emphasis on quality of service) and
IP (with emphasis on simplicity). In particular,
we have shown that multiple LSPs between
ELRs which separate not only O-D pairs but
also service classes in most realistic cases are
likely to be the preferred mode of operation.
We next considered path selection and band-
width allocation in multi-service MPLS net-
works in order to optimise the network quality
of service. The optimisation was based upon the
constrained optimisation of non-linear objective
functions. Two examples of such functions were
shown in detail, an Erlang-B based one for ser-
vices carried by e.g. UDP and where equivalent
bandwidth applies and a CAC mechanism is
active, and another one based on best effort ser-
vices over TCP where the transmission rate is
adapted to the level of congestion. We also dis-
cussed the possibilities to include other functions
to suit, e.g., services carried by UDP but for
which CAC does not apply.We presented a computationally efficient algo-
rithm called XFG to find and capacitate optimal
LSPs. The algorithm is based on a bandwidth
market where bandwidth prices determine the
routes and bandwidths of LSPs. The algorithm
was applied to compute optimal LSPs for a 55
node network model carrying 6 service classes
and for an 8 node network carrying 2 service
classes. It was seen that the method is capable
of quickly designing complex networks of LSPs
with appropriate quality of service discrimina-
tion between different service classes and that
also accounts for node performance characteris-
tics.A UMTS core network was used as a concrete
example to show how the method can be applied
to design general multi-service networks, and
the extension to include VPNs was described.The method can be rephrased to distributed
design. A first approach is outlined in [8] and
further work is in progress. Other points of study
include cost based pricing to model design of
MPLS networks on leased lines. Finally, work
on a more elaborate TCP model which includes
short file transfers, slow start, time-outs and a
more advanced queuing model is currently in
progress.References
1 IETF. Rosen, E et al. Multiprotocol Label
Switching Architecture.2001. (RFC 3031.)2 IETF. Awduche, D et al. Requirements for
Traffic Engineering over MPLS.1999. (RFC
2702.)3 IETF. Andersson, L et al. LDP Specification.- (RFC 3036.)
4 Aboul-Magd, O et al. Constraint-Based LSP
Setup Using LDP.2001. IETF draft.5 Ashwood-Smith, P et al. Generalized MPLS
Signaling – RSVP-TE Extensions.2001.
IETF draft.6 IETF. Davie, B et al. MPLS using LDP and
ATM VC Switching.2001. (RFC 3035.)