Hi,

Christopher Morrow wrote:
> My feeling is this:
> hop-by-hop processing will happen in slow-path (if you permit it to
> happen at all), you can't build a router today with an asic that'll
> know how to handle options which are created in the future :(

while the latter is surely true, I don't think that this
argument precludes to implement fast path processing
of at least _some_ HbH options. I think that the current
implemented strategy to simply pass all HbH options
to the slow path CPU is broken.

I'm not sure whether it is feasible, but couldn't the
Fast Path (ASIC) at least check for:
- known options to be processed in the fast path, e.g.,
  some CONEX info field
- known options to be processed in the control plane,
  i.e., punt packet to control path (slow path CPU),
  e.g., some to intercept packets with a specific
  router alert option
- bypassing unknown options (depending on the specified
  actions) in the fast-path

So some (not all) _basic_ header extension processing code
should be down in the fast path (like forwarding of non-supported
or "non-interesting" hop-by-hop options) while more complex
processing can surely stay up in the slow path.
I have the impression that we will sacrifice the possibility of
future useful IPv6 extensions due to some existing
implementation limitations.

Regards,
 Roland

--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------

Reply via email to