Re: [Idr] community of the day - common header

[Jeffrey Haas <[e-mail]>]

On Fri, Sep 09, 2016 at 11:44:13AM -0400, Jared Mauch wrote:
> > On Sep 9, 2016, at 11:33 AM, Jeffrey Haas <[e-mail]> wrote:
> > We know these things are community-like.  Stick them under a common type and
> > let people's policy engines filter them.
> 
> Sure, implementations compliant with 7606 should do
> treat-as-withdrawl if they encounter something they
> can recover from.

Right, but the filtering mechanism you're familiar with (which we strongly
recommend against as common practice) means you're applying the filter
potentially on nodes that are ignorant of internal contents.  I.e. you're
filtering - and my point about the common filter is to permit filtering.

> > This thread isn't about wide communities.  Stop making it about that.
> 
> Well, is it about rate of consumption of attributes, any new
> route coloring/community-type thing?  It’s clear from me watching
> the live stream of operator events today account teams are going to face
> a number of requests about implementation details.

It's about providing a filtering knob that doesn't go into throwing out
entire path attribute sets.

> Please don’t trim too much of my operational grouse here, we should seek
> consensus on *something* without prescribing what it is, wide, jumbo,
> skinny or otherwise.  We as operators need something, i’m less worried
> about the wire encoding than a viable solution.

I'm the one that's agnostic on wire encoding and not allergic to TLVs. :-)

Turned around, this is mostly about providing an operational tool.  A big
portion of the extended discussions not only at this last IETF but among
authors of wide and the older flex comms proposal was the fact that once
format flexibility becomes easy, we're likely to see a plethora of proposals
covering local use cases.  The push for BGP as a general signaling mechanism
for non ISP uses has been a large motivator there.

As one example, Enke's geo-tagging for routes was proposed as yet another
attribute.  It fits into a "community" bucket just as readily.

As an operator, do you want to play long-term whack-a-mole with the
attribute set of the week, worrying about whether something you filter today
may need to get through later, or would you prefer we make that job easier
but trying to move "color" into a common bucket?

-- Jeff