Re: [tsvwg] RTT-independence

[<[e-mail]>]

Bob,

your argument to the point below is valid. There’s a “working point”, to which a queue-depth (and AQM, if present) of a network provider access router is configured. If conditions significantly differ from those assumed for the working point, either transport throughput for all or a share of flows will deteriorate (or buffer might be underutilized).

I’d appreciate if test buffer depths, AQM configuration (if present) and RTTs are chosen reasonable to the above regard. That doesn’t mean, RTTs or which aren’t prevalent today shouldn’t be considered. But RTTs should be picked or mixed so that the overall test design including queue-configuration is operating in a mode to be expected with such a working point in mind.

I don’t volunteer to pick one of these advanced test scenarios. I note that 5G comes with partially low RTT expectations (10 ms in the most demanding case, I guess). I’m not a wireless network or wireless architecture expert however.

Regards,

Ruediger

<snip>





The L4S proponents argue, that this is working as designed, as this is a side-effect of their under-explained choices of acceptable standing queue "targets" for the two queues (1ms for the L4S-queue and 15ms for the non-L4S queue), which for short path RTTs in the limit creates an RTT imbalance of 1:15. (I have also argued that 15ms is the wrong value in theoretical grounds and so far this has not been disproven with facts/data).
15ms comes from the amplitude of the sawteeth of typical traffic at typical Internet bottlenecks over typical paths. "Traffic" might be multiple flows or single flows, but an AQM has to work well for the cases where the user wants to use the bandwidth they are paying for, including for single long-lived flows (OS updates, video uploads, etc). "Traffic" might use Cubic, Reno or other CCs but the user doesn't control the CC the other end uses, so the AQM has to work for typical cases and not be awful for worst cases. "Bottlenecks" might be anywhere, but they are most often in access links where there is not much flow multiplexing. With Reno, Cubic etc, the sawtooth amplitude depends on the RTT of the path. So "paths" might be between a customer in a large urban conurbation and a CDN in the same or a nearby conurbation, or a path might be half way round the globe, but the common case these days is an urban customer and a CDN.

One can argue whether the number is 10ms, 15ms, 20ms or whatever, but there is an operating point below which more and more traffic badly underutilizes the link. Whatever, this is not a theoretical question, it is a commercial question for ISPs when setting the target parameter of their AQM. You have to look at what ISPs do, not what you think they should do.