Re: [Detnet] Magnus Westerlund's Discuss on draft-ietf-detnet-mpls-over-udp-ip-07: (with DISCUSS)

[Stewart Bryant <[e-mail]>]

Dn needs to run in a well managed network to fulfil its loss delay obligations. If the quality of the network is such that the packets may get corrupted then that will have as significant an impact on the DN service, because failure to deliver is as significant an issue as failure to deliver correct packets. In other words whilst the UDP c/s will save the receiving node from receiving a corrupt packet, that failure to deliver also causes applications issues. The only satisfactory way to address this is to ensure the quality of the network is such that rate of corruption is trivial, and thus the rate of c/s error detection would  also be trivial.

It should also be noted that most routers that are capable of high-speed low jitter packet deliver do not have the h/w capability to compute a whole packet c/s on the fly and hence would have to punt the packet to the slow path, which in turn is contra to the requirements of deterministic delivery of the packet.

- Stewart

> On 9 Dec 2020, at 22:53, Black, David <[e-mail]> wrote:
> 
> Suggestion ...
> 
> Stewart - could you write a brief explanation of this statement (from your email to the list)?:
> 
>>> We agree that Detnet fulfils the requirements to run [over IPv6] with no [UDP] c/s,
> 
> The rest of us could then look at that explanation and figure out whether the situation is "always fulfills" (would be nice if true) or "almost always fulfills" (might have to deal with that).
> 
> Thanks, —David