[LOOPS] Narrowing the scope: (3) Underlay (segment) path assumptions and ECN

[Carsten Bormann <[e-mail]>]

In tomorrow’s LOOPS side meeting, we want to discuss whether we can narrow the initial scope of the WG sufficiently that the resulting charter looks more tractable (we can then recharter after delivering on the initial scope).

LOOPS recovers packets that are lost on its path segment.  That counteracts the effect of packets dropped by that segment to slow down traffic during congestion.  Preferably, we would find ourselves be able to assume that all packet losses are not congestion related.  A fully ECN-capable path with ECT(0) traffic on it would satisfy this requirement to a large extent — routers would signal congestion by CE-marking.

Should we assume segment paths built from routers that are fully ECN enabled (i.e., indicate congestion by CE-marks)?
(Note that ascertaining this would be under control of the operator of the LOOPS pair, who might want to take decidedly non-bottleneck routers out of the consideration to simplify their job.)

We probably still need some circuit-breaker mechanism in case the path does seriously degrade and needs to drop packets for congestion reasons.
We also would want to limit the effect of wrong attribution of losses by an operator-setup overall limit to the overhead added by LOOPS.

Comments are welcome on the list and/or in the meeting in ~16 hours.

Grüße, Carsten