Re: [lisp] Suggested UDP checksum text

[Sam Hartman <hartmans-ietf at mit.edu>]

>>>>> "Noel" == Noel Chiappa <jnc at mercury.lcs.mit.edu> writes:

    >> From: Sam Hartman <hartmans-ietf at mit.edu> I think we need to at
    >> least say that for LISP data plane packets, an ITR MAY send a 0
    >> checksum.

    Noel> We may want to upgrade that to a "SHOULD"; my reasoning for
    Noel> this is a cost-benefit analysis. Furthermore, depending on
    Noel> how we word the rest of the statement (i.e. covering
    Noel> reception processing), it may actually be very desirable to
    Noel> suppress generation of checksums.

    Noel> One has to start with the recognition that this non-end-end
    Noel> checksum buys us very little, simply because it's
    Noel> non-end-end. (In fact, if we have applications which have an
    Noel> actual use for damaged packets, it will have a _negative_
    Noel> impact, as it will cause such packets to be discarded.)


I think that if I thought about it  enough I'd find the above compelling
and end up at:

* senders SHOULD NOT send a checksum
* Receivers SHOULD check a checksum if present

I'd be using your reasoning on the send side.  My reasoning on the
receive side is that changing the UDP handling seems problematic to me
and unjustified especially if we are recommending that senders not
send checksums.

However, if we can all live with MAY/MAY, I think we'll be mostly done
with this issue, so while I'm happy to continue circling if people
really want to,  we could just declare victory on MAY/MAY.

The SHOULD NOT/SHOULD comment is my personal opinion.  However I
believe we're close to consensus on MAY/MAY as a chair.  Also, as a
chair, I'm not seeing any new information added to this discussion at
this point.