Re: [Bier] comments for draft-zhang-bier-bierin6-04.txt

[Toerless Eckert <[e-mail]>]

On Tue, Mar 17, 2020 at 02:00:15PM +0000, IJsbrand Wijnands (iwijnand) wrote:
> Hi Toerless,
> 
> Picking up on this thread, little late, sorry..
> 
> > I liked Pierre's bitstring in IPv6 addr, i just felt it was too short to be useful.
> 
> It would disagree, it???s about half the size of what we consider to be the standard interoperable mask of 256. We have defined the Set approach to overcome BM length limitations, at the expense of more copies sent by the ingress. Pun intended, BIER-TE is much more restricted to be useful ????

No pun needed. Valid point:

With BIER-TE, the goal is to get tree engineering with bitstrings, 
so we can get per-packet trees, and we think there will be good use-cases
too in more constrained environments, such as broadband access (see
our http draft) as well as in 'DetNet' type manufacturing and the like,
where hopefully the forwarders (due to lower speeds than 100Gbps++)
would be easier to support longer bitstrings as needed. And the length
of bitstrings is not nailed in stone forever like in IPv6, so
different markets could evolve to support different (longer) length.
The 256 is really todays lowest common denominator that we can
support in all SP edge/core-routers.

With IPv6 via bitstrings in addresses (and correct me if you disagree),
the main attraction was the ability of letting applications do BIER
without having change APIs/host-stacks. That is cool short-term, but
the long-term benefit deteriorates as soon as someone does write
a BIER SDK for linux and windows. And with that context in mind i
found the long-term limitation of not being able to grow bitstrings
in various markets to be too limiting.

Cheers
    Toerless

> 
> Thx,
> 
> Ice.