[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: failure detection



On Wed, 17 Aug 2005, marcelo bagnulo braun wrote:

FWIW, TCP for instance already provide some form of such signaling see Appendix E.1 of RFC 2461 Of course other ULP don't provide any kind of feedback.

So i guess it would be a better not to fully rely on ULP feedback, but otoh not using the information that some ULP provide seems suboptimal to me.

1. I said "extra signaling"

2. Note the title of appendix E: "IMPLEMENTATION ISSUES" and
   describes a possible optimisation that can be made if ULP's can be
   modified - it's wholly at the implementation's discretion.

Exploring stuff like path-reachability probing or implementation optimisations via shim<->ULP signalling (which ULP could be IP quite likely btw, not TCP - implementations possibly will have /two/ layers to pass hints through) is fine, but do it as an appendix or an informational document or wait for implementors to experiment.

The path-probing is, imho, mostly a complete waste of time. But an implementation can go wild if it wants.

The potentials of TCP<->Shim signaling are interesting I'll admit. However, again it's an implementation detail (and only reason its being discussed is cause of the crazy path-probing stuff).

I note from the archives that I'm not the first person to have cautioned against going down the path-probing route (Joe Abley too).

regards,
--
Paul Jakma	paul@clubi.ie	paul@jakma.org	Key ID: 64A2FF6A
Fortune:
Manly's Maxim:
	Logic is a systematic method of coming to the wrong conclusion
	with confidence.