traversed routers. According to [RFC2702], the
advantages of MPLS can be further described by:
- Explicit label switched paths which are not
constrained by the destination based forward-
ing paradigm can easily be created through
administrative action or through automated
actions by the underlying protocols. - LSPs can potentially be efficiently main-
tained. - Traffic trunks can be instantiated and mapped
onto LSPs. - The attributes can be associated with traffic
trunks that modulate their behavioural charac-
teristics. - A set of attributes can be associated with re-
sources that constrain the placement of LSPs
and traffic trunks across them. - MPLS allows for both traffic aggregation and
disaggregation whereas classical destination
only based IP forwarding permits only aggre-
gation. - It is relatively easy to carry out constraint-
based routing with MPLS. - A good implementation of MPLS can offer
lower overhead than competing alternatives
for Traffic Engineering.
One way of designing MPLS networks is to
apply constraint-based routing. Then the follow-
ing information is commonly used as input:
- Attributes associated with traffic trunks;
- Attributes associated with resources;
- Other topology state information.
Based on this every node may find an explicit
route for each traffic trunk originating from that
node. Here, an explicit route for each traffic
trunk is a specification of an LSP that satisfies
the demand requirements expressed by the traf-
fic trunk’s attributes, subject to constraints im-
posed by resource availability, administrative
policy, etc.
A heuristic approach might follow two steps;
i) prone resources that do not satisfy the require-
ments of the traffic trunk attributes; and ii) run
a shortest path algorithm for the residual graph.
In general, when multiple traffic trunks are to
be routed, it cannot be shown that the algorithm
always finds a better mapping (or even a solution).
6.3 MPLS-support of DiffServ
Utilising MPLS to carry DiffServ classes has
been looked upon with much interest. This is
described in e.g. [ID_MPLSdiff]. Then the ques-
tion arises how to map the behaviour aggregates
(BAs) onto LSPs. Basically this can be done by
either:
- Using LSPs that carry several Ordered Aggre-
gates (OAs), implying that the Exp field in the
MPLS header is used for separating different
classes (giving scheduling treatment, prece-
dence, etc.). This is called Exp-inferred PSC
LSPs (E-LSPs). Then up to eight BAs (3 bits
in the Exp field) can be carried by a single
LSP. Mapping from Exp to PHB (PSC and
drop precedence) for an LSP is either explic-
itly signalled or pre-configured; or - Using LSPs to carry a single OA, saying that
the packet treatment can be derived from the
Label field while precedence might be derived
from the Exp field. This is called Label-only
inferred PSC LSPs (L-LSPs). Then an LSP
carries a single (FEC, OA) pair. The PSC can
be inferred from the label without looking at
other information. This implies that the PSC
is explicitly signalled at label establishment.
The Exp field may give the drop precedence.
In case ATM is used, information in the ATM
header can be used, e.g. the CLP field.
As mentioned above, combining DiffServ and
MPLS allows further differentiation, e.g. by
defining different levels of protection for the
LSPs.
DiffServ implies service differentiation at every
hop, while MPLS and TE may achieve better
traffic distribution of the aggregated traffic
loads. In this respect, they may work partly inde-
pendent of each other. Then, MPLS can apply
constraint-based routing and admission control
for all traffic flows carried in the same LSP. In
case more fine-tuning of resources and traffic
flows is sought, TE mechanisms may be applied
on each service class or smaller groups of traffic
flows, e.g. by mapping more specific traffic
trunks onto the LSPs.
Some requirements for support of DiffServ
by MPLS traffic engineering are listed in
[ID_DMreq]:
- Compatibility between mechanisms applied
on DiffServ and MPLS level; - Support of separate constraints on bandwidth
for the different classes; - No additional constraints on the number of
class-types and classes;