The MPLS WG Archive

[Curtis Villamizar <curtis@avici.com>]

In message <336ECDAFDF7DD311B9E30090277AEE4101B40623@nt-exchange-bby.pmc-sierra
.bc.ca>, Shahram Davari writes:
> If you are talking about per-LSP BW reservation, I don't consider that an
> MPLS QoS. This QoS is essentially either per-flow Intserv QoS, or
> per-flow-signaled aggregate Diffserv QoS.
> 
> In other words MPLS signaling (RSVP-TE, CR-LDP) gives you the ability to
> signal BW, but there are other protocol layers that provide the QoS, not
> MPLS. In other words MPLS in orthogonal to QoS but it facilitates providing
> QoS.
> 
> -Shahram


Shahram,

You are making very strong statements regarding features that you
claim that MPLS doesn't have.

MPLS TE, using draft-ietf-mpls-diff-ext-05.txt, provides a means to
signal traffic classifications that will later be applied on an MPLS
label and EXP basis to traffic.  The selection of PHB allows the
network operator to map aggregates of traffic identified by label and
EXP when classified as MPLS, into separate queues (for each BA, not
for each MPLS flow, unless a separate PSC value per flow is used).

At the ingress traffic can be classified by any means already defined
for IP (or whatever L3 is used, if not IP).  For IP this is expeced to
usually include DSCP, and may also include ingress interface or
ingress virtual interface.  In principle filters based on IP addresses
and port numbers or per microflow RSVP can be used on the ingress can
be used for classification, though classification at this fine a
granularity is expected to occur close to the edges of major networks
(if at all) and not in major backbones.

The traffic classification at ingress is used to select the tunnel
that provides the appropriate QoS.

The bandwidth reservation amounts allow such things as queue weights
to be set.  The aggregation of tunnels within tunnels allows the sum
of bandwidth within an outer tunnel to be reflected in the reservation
for the outer tunnel.

Although so far bandwidths on MPLS tunnels have been based on
configured values based on historic data, bandwidths could also be set
based on (filtered) measurements of traffic.

The statement that MPLS is orthogonal to QoS is simply not true.

Curtis