Re: [tsvwg] Comments on draft-ietf-tsvwg-udp-options-07

[Tom Herbert <[e-mail]>]

On Mon, Mar 11, 2019 at 7:54 PM Joe Touch <[e-mail]> wrote:
>
>
>
> > On Mar 11, 2019, at 7:24 PM, Tom Herbert <[e-mail]> wrote:
> >
> > Considering that this is an application based mechanism, how would the
> > average user even know to care about this?
>
> The same way they know and care about udp checksums or not.
>
> > It's better to not give the
> > option and just provide something that is robust out of the box.
>
> We already have plenty of choices that are more risky, including CS=0.
>
That's more an argument that UDP checksums should not be zero, rather
than a fixed checksum is not needed in UDP surplus space. I agree with
that. Given that every NIC offloads the checksum anyway, the
recommendation really should be to always use the UDP checksum (it's
required in IPv6 anyway and we get better performance in cases of
encapsulation).

> The point is to be flexible - were not designing for today only.
>
Robustness trumps flexibility.

> And space - even a few bytes - is absolutely a significant consideration.
>
It depends on several factors whether the overhead is justified. For
instance, a four byte alignment requirement really does have any
impact since IP and UDP are already four byte aligned and MTUs are
almost always divisible by four (i.e. alignment doesn't reduce
available MTU). If the common deployment case is to always enable OCS
(seems like in Appendix A of the UDP options draft the intended
default is to enable it), then the overhead of the header reduces to
one byte. An overhead of one byte really is not worth discussion.

Tom

> Joe