[LOOPS] How important is congestion detection

[Carsten Bormann <[e-mail]>]

After the BOF, I had an interesting hallway conversation with Bob Briscoe on congestion feedback signaling, and about congestion detection (i.e., grouping losses into non-congestive and congestive).

One assumption behind the solutions strawman draft was that it is a valid approach to signal loss events that occur inside the path segment, as CE markings to the end hosts.  Bob pointed out that some end host implementations today interpret CE as definite congestion events, while applying congestion detection to loss signals, so this conversion might be counterproductive.  Congestion detection is then a necessary component of a LOOPS implementation, in order not to be worse than the congestion detection employed by the end hosts.

Congestion detection does not need to be standardized (we probably do want to have a BCP or two at some point, though), so that would not be a theoretical roadblock.

But would it be, in a practical sense?  Do we expect congestion detection along a full path (as Stuart Card said today, against the composition of the statistical behaviors of multiple segments) to be better than in a LOOPS implementation looking at a single path segment?

Grüße, Carsten