Re: [Pce] Benjamin Kaduk's Discuss on draft-ietf-pce-gmpls-pcep-extensions-14: (with DISCUSS and COMMENT)

[Benjamin Kaduk <[e-mail]>]

On Mon, Jul 01, 2019 at 11:45:04AM +0200, [e-mail] wrote:
> Hi Ben,
> 
> Thank you for following up. Please see [JM] below.
> 
> 
> On 29/06/2019 02:51, Benjamin Kaduk wrote:
> > Hi Julien,
> >
> > On Fri, Jun 28, 2019 at 03:26:33PM +0200, [e-mail] wrote:
> >> Hi Ben,
> >>
> >> Apologies for the late response.
> > I probably wouldn't have been able to reply much before now anyway -- I've
> > been quite busy (and have some other documents' authors waiting for
> > feedback as well).
> >
> >> Let us focus on your first DISCUSS comment which is the main one. You
> >> raised a legitimate point that we acknowledge. After discussion, we
> >> propose to modify the end of section 2.4 as follows:
> >>
> >> OLD
> >>
> >>    The semantic of the LOAD-BALANCING object is not changed.  If a PCC
> >>    requests the computation of a set of TE-LSPs so that the total of
> >>    their generalized bandwidth is X, the maximum number of TE-LSPs is N,
> >>    and each TE-LSP must at least have a bandwidth of B, it inserts a
> >>    BANDWIDTH object specifying X as the required bandwidth and a LOAD-
> >>    BALANCING object with the Max-LSP and Min Bandwidth Spec fields set
> >>    to N and B, respectively.
> >>
> >> NEW
> >>
> >>    The semantic of the LOAD-BALANCING object is not changed.  When the
> >>    bandwidth can be expressed as scalars (e.g., for circuit-switched
> >>    technologies), if a PCC requests the computation of a set of TE-LSPs
> >>    so that the total of their generalized bandwidth is X, the maximum
> >>    number of TE-LSPs is N, and each TE-LSP must at least have a
> >>    bandwidth of B, it inserts a BANDWIDTH object specifying X as the
> >>    required bandwidth and a LOAD-BALANCING object with the Max-LSP and
> >>    Min Bandwidth Spec fields set to N and B, respectively.
> >>
> >>    For technologies based on statistical multiplexing (e.g., InterServ,
> >>    Ethernet), the exact bandwidth management (e.g., Ethernet's Excessive
> >>    Rate) is left to the PCE's policies, according to the operator's
> >>    configuration. If required, further documents may introduce a new
> >>    mechanism to finely express complex load balancing policies within
> >>    PCEP.
> >>
> >>
> >> Would that update address that part of your DISCUSS? (The remaining
> >> parts will be straightforward once this one is resolved.)
> > I seem to have forgotten some of the details of what fields are in the
> > various objects, but let's see what I can say regardless of that.
> > I think the general approach of specifying behavior when there's an obvious
> > scalar bandwidth and leaving the choice to the PCE in other cases is a good
> > approach.
> >
> > For the specifics, we probably don't want to say "when the bandwidth can be
> > expressed as a scalar" and "for technologies based on statistical
> > multiplexing" just in case there ends up being some third category, so the
> > second paragraph could start off as something like "In all other cases,
> > including for technologies based on statistical multiplexing [...]".
> [JM] You're right. That's a safer wording, easy to include.
> 
> > It also seems like for the first case, we would be able to use the
> > un-modified RFC 5440 BANDWIDTH and not need the generalized form?  But
> > perhaps I misremember/misunderstand...
> [JM] This is indeed what the current wording suggests, so I propose to
> change "expressed by scalars" with "summarized into scalars". The idea
> here is to accommodate existing constructs from RSVP-TE signaling, which
> are compound objects allowing to specify more detailed information than
> aggregated values. E.g., a single 10 Gb/s circuit vs. a group of 4 x 2.5
> Gb/s containers.
> That would lead us to the following text:
> 
>    The semantic of the LOAD-BALANCING object is not changed.  When the
>    bandwidth can be summarized into scalars (e.g., for circuit-switched
>    technologies), if a PCC requests the computation of a set of TE-LSPs
>    so that the total of their generalized bandwidth is X, the maximum
>    number of TE-LSPs is N, and each TE-LSP must at least have a
>    bandwidth of B, it inserts a BANDWIDTH object specifying X as the
>    required bandwidth and a LOAD-BALANCING object with the Max-LSP and
>    Min Bandwidth Spec fields set to N and B, respectively.
> 
>    In all other cases, including for technologies based on statistical
>    multiplexing (e.g., InterServ, Ethernet), the exact bandwidth
>    management (e.g., Ethernet's Excessive Rate) is left to the PCE's
>    policies, according to the operator's configuration. If required,
>    further documents may introduce a new mechanism to finely express
>    complex load balancing policies within PCEP.
> 
> Is it better?

That is better; thank you.

The only other comment I have would be to ask if it's worth saying
something in the first paragraph about how "other link-technology-specific
parameters are not included in the bandwidth calculation" (with my
apologies for probably using the wrong terminology; this is not my area of
expertise).

-Ben