Re: [tsvwg] UDP Options: on forcing the use of UDP CS=0 in connection with FRAG+LITE

[Tom Herbert <[e-mail]>]

On Mon, Jul 1, 2019 at 1:19 PM Joe Touch <[e-mail]> wrote:
>
>
>
> > On Jul 1, 2019, at 12:49 PM, C. M. Heard <[e-mail]> wrote:
> >
> >> You haven’t addressed the data copying issue.
> >
> > Yes, I have, in
> > https://mailarchive.ietf.org/arch/msg/tsvwg/RJBWi_tiW6M_phfc22zKuVn_vh4
> >
> > Can you please respond to that message and explain exactly why you think
> > that the proposal in the thread "DP Options: how to do FRAG without LITE
> > and forced UDP CS=0" imposes data movement requirements for packet
> > reassembly beyond those required by FRAG+LITE (or FRAG without LITE) in
> > in draft-ietf-tsvwg-udp-options-07? I sure don't see it.
>
> When I’m done moving a small, fixed number of bytes, the existing FRAG+LITE option leaves the fragment at the head of the packet.
>
> When you’re done, the frag is after a variable length string of options.
>
> Yes, both need to be gathered together. No, they’re not the same - some UDP hardware and software implementations treat the offset where they expect the data to start as “special”; yours could be anywhere.

Joe,

If we need to perform the reassembly checksum in the host CPU then any
performance gains your seeing trying to keep the data early in the
packet would be eclipsed by all the cache misses we'll be taking for
the calculation. We have a much better chance to be able to offload
the checksum if it's per packet instead of being over the reassembled
packet (the merits of per packet checksums have been previously
discussed).

Tom

>
> Joe