heistp
commented
9 months ago The risk is more serious than just from deployed single queue RFC 3168 AQMs though. The Likelihood section documents that even when FQ is deployed (i.e. fq_codel, the default in many Linux OSs, and on many WiFi routers), L4S and non-L4S traffic can end up in a shared RFC 3168 signaling queue when they're in the same tunnel, or wind up in the same queue due to a hash collision. Most traffic in the same tunnel shares the same 5-tuple hash, and will thus end up in the same RFC3168 signaling queue, in the case of fq_codel or Cake, for example.
Unless this is tested for explicitly, it may well go unnoticed, or rather "unattributed", as when a problem occurs that affects their QoE in some way, it's unlikely that the user will be able to identify this as the reason.
A risk is documented a bit disingenuously. A key risk is about single-queue RFC3168 AQMs but the document says there is no info on any such deployments - but in too-neutral of a way that may mislead a reader. While highlighting the risk you may wish to note that it is largely theoretical given no one is aware or any scale deployment of single-queue RFC3168 AQMs. If you are aware of those, then they should be documented and the volume of traffic relative to the total volume of traffic should be cited.