Re: [arch-d] ETSI launches new group on Non-IP Networking addressing 5G new services

[Toerless Eckert <[e-mail]>]

On Thu, Apr 09, 2020 at 02:35:21PM -0700, Tony Li wrote:
> > Aka: IMHO the argument of variable length addresses is still 
> > quite an open issue..
> 
> 
> It shouldn???t be.  Today???s hardware is enormously capable and continues to grow aggressively (thank you Gordon Moore).

Nice to hear. I was thinking this for a long time for higher end
platforms, but i am unsure about TCAM architectures with fixed
upfront parsers. No ideas about the latest lower end, e.g.:
enterprise level switches. P4 also seems to have overheads
with variable length parsing (only hearsay, need more cycles
to read into those levels of P4 details).

> It would be backwards to let hardware limitations constrain the architecture.  Instead, the hardware will always grow to what is needed by the architecture.

Did you watch the forwarding plane reality talk we gave a few IETFs ago ?
There was still a lot of concern about the right degree of flexibility to
expect from one of the co-presenters slides.

> That said, that???s no excuse to create enormously wasteful architectures.  Resources are not free. Waste not, want not.

Well... I had discussions with a bit Service Provider wondering about what
they thought about the overhead of e.g.: 6x128bit SRv6 SIDs, and he
was not concerned. One person of course, but i think we have to assume
that there is a wide range of opinions about what would constitute
"wasteful" on the wire. 

Variability in headers is a nice concept to make more use-cases happy
IMHO, but when can we really expect it to be acceptable on any
low-end switch (worst hardware).

> 
> Tony

-- 
---
[e-mail]