Re: [Idr] WG adoption of draft-heitz-idr-large-community; one week to comment on early code point allocation

[Jeffrey Haas <[e-mail]>]

> On Sep 27, 2016, at 4:24 AM, Job Snijders <[e-mail]> wrote:
> 
> On Mon, Sep 26, 2016 at 07:31:57PM -0400, Jeffrey Haas wrote:
>> FWIW, I think things with "well known" semantics should stick to the
>> reserved code point space in 1997 communities unless there's some
>> motivation to have a common parameter/argument in it.  In that case,
>> it fits the case for large comms anyway.
> 
> I agree with you Jeff. 
> 
> How about the following:
> 
> """
>    The Large Community attribute values in the following ranges are
>    reserved:
> 
>             0:0:0 -          0:4294967295:4294967295
>    4294967295:0:0 - 4294967295:4294967295:4294967295
> 
> """
> 
> This way the lowest and highest possible value are reserved, and at some
> later point (part of) these reserved ranges could be converted to
> Well-known or something else if that need arises.

I'm fine with such a reservation.  However, it'd be good to get the reasoning in the doc rather than letting it be inferred after the fact.

As one example, the "don't use AS0" seems like a reasonable thing since it's not permitted in the Internet, but I'm aware of provider policies that are using 0:* in 1997 format to signal stuff.  Arguably that case gets weaker with the additional field in -large.

The 65535:* in 1997 was fine since something had to be carved out for the well knowns.  Where this has been awkward has been the tendency of providers wanting to actually *use* 65535 as a private AS number, but now having an overloaded community space for what's being used as an AS.

> 
> Going back to my earlier question, reserving 65535:*:*, just because it
> was a meaningful value in rfc1997, does not make a whole lot of sense.
> 65535 was picked in rfc1997 because it was the highest value, we can
> apply the same logic with Large.

I tend to agree.  This piece of the conversation mostly needs harmonization with "what do we do or don't do about automatically mapping 1997 into -large".  I'm agnostic on that point.

> 
> Note: i propose just to reserve these two ranges, I don't see an
> immediate need to open a registry given that we already have rfc1997
> WKC.

I concur, modulo the above.

-- Jeff