Hello,
We agreed in
Vienna that the relationship between the Shared Mesh Restoration mechanisms
being developed in CCAMP and the DS-TE Bandwidth Constraints Models needed to
be investigated so we can decide "if there is or is not an issue and if
it needs to be addressed or not".
In a nutshell (see
[A] below for more details), our conclusions are
that:
- Shared Mesh Restoration can work simultaneously with DS-TE.
- Shared Mesh Restoration should operate independently within each DS-TE
Class-Type (and not across Class-Types).
- Shared Mesh Restoration can
work with RDM, MAM and MAR
Thus, our proposal
for moving ahead on this is:
- make a wording change to the definition of "Reserved (CTc)" which is used in
the formulas for defining RDM, MAM and MAR so that the formulas are compatible
with how Shared-Mesh Restoration performs bandwidth reservation/CAC (see [B]
below for exact word change)
- add a note in RDM, MAM and MAR specs that these BC Model definitions
are compatible with Shared Mesh Restoration with the assumption that
Shared Mesh Restoration operates independently within each
Class-Type.
I would like to
thank Yakov, Anna and JP for working with me on this
topic.
Please let us know
promptly if you have comments/issues with this.
Thanks
Francois
[A]
===
Shared Mesh
Restoration is defined in draft-ietf-ccamp-gmpls-recovery-functional-00.txt
(see section 3.3).
One key concept is
that backup LSPs can share bandwidth as long as they protect from failure of
different resources (since signaling is used at failure time to activate a
particular pre-established backup LSP). Because they can share bandwidth, it
is clear that the total amount of bandwidth actually *reserved* by the backup
LSPs can be smaller than the sum of the bandwidth *signaled* by each
individual backup LSP.
Another concept
discussed in the context of Shared Mesh Restoration is "Excess Traffic
LSP" (altough I think this is not yet documented in the current draft but will
be added in upcoming one). Excess Traffic LSPs may be
established (at a lower preemption priority) and use the resources allocated
(but not currently used) by backup LSPs (at higher preemption priority). The
idea being that these Excess Traffic LSPs will get preempted as soon as
resources are actually needed by some backup LSPs. Since Excess Traffic
LSPs "borrow" bandwidth from backup LSPs when those don't need it, it is clear
that the bandwidth actually "reserved* collectively by backup LSPs and Excess
Traffic LSPs is smaller than the sum of the bandwidth *signaled* separately by
backup LSPs and by Excess Traffic LSPs.
So a first
conclusion is that the definition of "Reserved (CTc)" needs to be ajusted
to take into account the fact that the amount of bandwidth actually reserved
collectively by LSPs is not just a sum of bandwidth reserved individually by
all LSPs but rather that multiple LSPs may share bandwidth resources and
that what matters is the amount of bandwidth *actually reserved" across the
set of LSPs.
So now, how
can Shared Mesh Restoration operate in conjunction
with DS-TE?
We recommend that
Shared Mesh Restoration operates only within each Class-Type. This means that
bandwidth sharing across Primary LSPs/Backup LSPs and Excess Traffic LSPs only
occurs within each given CT. In other words, back up LSPs protecting
Primary LSP of CTx are expected to also belong to CTx. Similarly Excess
Traffic LSPs sharing bandwidth with Backup LSPs of CTx are expected to also
belong to CTx.
In this model, we
can see that :
- Shared Mesh Restoration model will define how much bandwidth is actually
collectively reserved by Primary LSPs + Backup LSPs + Excess Traffic
LSPs within a given Class-Type.
- DS-TE Bandwidth Constraints Model defines the set of bandwidth constraints
applicable to the bandwidth collectively reserved by a given
CT.
- the two models (ie Shared Mesh Restoration and BC Models) are effectively
orthogonal; ie the definitions of the Shared Mesh Restoration concepts affect
how one computes the amount of bandwidth actually reserved by a given CT, but
this does not affect/modify which/how the bandwidth constraints apply to each
CT. Thus, Shared Mesh Restoration can operate simultaneously with DS-TE with
RDM, MAM or MAR BC models and these BC Models can be defined independently of
eth detailed concepts/rules of Shared Mesh Restoration.
Note that,
instead, one could conceive an extremely generic model for how Shared
Mesh Restoration operates with DS-TE whereby Shared Mesh Restoration spans
arbitrarily across Class-Types, for example where Primary LSPs (or
Excess Traffic LSPs) of one CT could share bandwidth with backup LSPs of
another CT. However, this does not seem to offer significant real practical
benefits while it introduces a significant level of complexity. For a start,
in such a model Sahred Mesh Restoration and BC Models are no longer orthogonal
and the definitions of Shared Mesh Restoration concepts may affect how
the bandwidth constraints affect each CT. We concluded that the
complexities of such a generic model were not justified and not worth
pursuing at this stage.
[B]
==
Current
definition is:
"Reserved
(CTc)" is defined as the sum of the bandwidth reserved by all established LSPs which belong to CTc.
New Definition
will be:
"Reserved (CTc)" is defined as the total
amount of the bandwidth reserved by all
established LSPs which belong to CTc.