Re: Operational feedback on policy redundancy

I think this is a VERY important point to make. One team might be
responsible for VOIP, so they set their intent requirement, which ignores
any other intent requirements coming from other teams. The
duplication/conflict is VERY intentional.

Shane

On Sat, Apr 4, 2026 at 3:17 PM Pedro Prado via NANOG <[email protected]>
wrote:

> My 2c...
> Your examples sound like decisions made in different contexts - perhaps of
> time, or focus.
> The VoIP requirement was probably set first since this is fundamental for
> it, and at a later time the overall latency requirement was set, but at a
> broader level. Or perhaps it is known that the general latency requirement
> isn't that "set in stone" and it's safer to keep a <25ms requirement at the
> service level for VoIP should the general requirement change.
> The same can be applied to ACLs: you might apply a protection that in
> principle should never be needed to an element because the other
> "protector" might still fail, be mishandled, etc. Probably even more so if
> it's an intent-based system - both the intent and the translation of the
> intent to the actual device's configuration might have issues that one
> might want to avoid being a victim of, in certain cases.
>
> In general, though... if everything is intent-based, in a very solid system
> intent redundancy and intent conflicts should be easily detectable and
> avoidable. If desired and justifiable since that does not come free...
>
> *Pedro Martins Prado*
> [email protected] / +353 83 036 1875 (FaceTime & WhatsApp)
>
>
> On Sat, 4 Apr 2026 at 19:46, Gary Sparkes via NANOG <[email protected]
> >
> wrote:
>
> > In my view, one or the other can be overridden, while the other can't.
> >
> > For example, that 'general' latency requirement, while ridiculous, could
> > theoretically be waived for services that are located off-site or in a
> > different geographic region. While the VOIP requirement is hard for
> service
> > delivery.
> >
> > Firewall rule interaction can also be a tricky space because behavior may
> > not appear evident from what the rules say 'in plain language' but how
> they
> > technically interact.
> >
> > You can have traffic traverse a firewall and just pass on without
> > processing, as well.
> >
> > Some also make some technical/functional sense too - IE: On BGP, have an
> > overarching announce of our entire /32 space, but also announce more
> > specific prefixes at the locations they're required. The more specific
> > prefix 'wins' in BGP, so it may appear redundant or confusing at first.
> >
> > Prioritization is definitely key, but as you see with the 20 vs 25ms
> > above, while the general policy might be waved, you might still have
> > specific sub requirements for specific things that would need to be
> waived
> > as well, or adhered to without *further* waivers/exceptions.
> >
> > -----Original Message-----
> > From: manwar--- via NANOG <[email protected]>
> > Sent: Saturday, April 4, 2026 2:28 PM
> > To: [email protected]
> > Cc: [email protected]
> > Subject: Re: Operational feedback on policy redundancy
> >
> > Hi all,
> >
> > Thanks for the feedback, and apologies if this isn’t the right forum for
> > this kind of question.
> >
> > To clarify: the data comes from an intent-based enterprise network, where
> > the intents are high-level requirements collected from a running
> production
> > system.
> >
> > By redundancy, I mean cases like:
> > - A general requirement (e.g., “latency < 20ms for all services”)
> > alongside a weaker, service-specific one (e.g., “VoIP latency < 25ms”),
> > where the latter is effectively subsumed.
> >
> > By conflicts, I mean situations like:
> > - One intent requiring all traffic to traverse a firewall, while another
> > requires no middleboxes for performance-sensitive services.
> >
> > In this dataset, such cases often appeared without explicit documentation
> > of how they were resolved. My assumption is that, in practice, these get
> > handled via implicit prioritization or later clarification.
> >
> > So my main question is: At the high-level goal / intent layer (before
> > translation into ACLs, BGP policy, etc.):
> > - Do redundant or overlapping requirements tend to exist in practice?
> > - Is it common for conflicts to be resolved through undocumented
> > clarifications or implicit prioritization?
> >
> > I do intend to publish the results of this work once the project is
> > complete, with the goal of making it useful for operators as well.
> >
> > Appreciate any insights.
> >
> > Best regards,
> > Mubashir
> > _______________________________________________
> > NANOG mailing list
> >
> >
> https://lists.nanog.org/archives/list/[email protected]/message/4JDEGK25VXD74NSLJXJVVFDCEZFXLSK6/
> > _______________________________________________
> > NANOG mailing list
> >
> >
> https://lists.nanog.org/archives/list/[email protected]/message/HGOC44MU5R43FYL3HWHKGJZOH6ZLUIND/
> _______________________________________________
> NANOG mailing list
>
> https://lists.nanog.org/archives/list/[email protected]/message/6TO3UN75ID4CLTUYMAGAKJ3OZESTK2UW/
_______________________________________________
NANOG mailing list 
https://lists.nanog.org/archives/list/[email protected]/message/QJFBKCCZXD2CNKOR6SIMVA6MALRZULTR/