Correct and update Prop324 based on things learned in prototyping

While testing congestion control over onion services, I noticed some omissions from the proposal that were present in the background material and literature, as well as some new heuristics I discovered from testing the prototype:

1. The congestion window only should be updated with a congestion signal once per window
2. If the local orconn is blocked, that should be an immediate congestion signal. Also doc that we may have too large a queue there
3. If the edge connections do not have data to send, estimates of BDP should not be updated
4. Westwood may have a runaway condition where max RTT continues to grow. We may want to reduce the max RTT measurement upon congestion
5. I made a congestion control algorithm that directly uses the current BDP estimate as its current congestion window, and this works. We should spec it and evaluate it in Shadow.
6. Update the consensus parameter list and tuning experiments section
7. BDP estimation algs

added S61-O3-Maybe - FINISHED Sponsor 61 - FINISHED labels

assigned to @mikeperry

@dgoulet - feel free to add anything else that you feel needs clarification

changed milestone to %Sponsor 61 - Making the Tor network faster & more reliable for users in Internet-repressive places

added Backlog label

marked this issue as related to #32 (closed)

changed the description

changed the description

WIP branch at https://gitlab.torproject.org/mikeperry/torspec/-/commits/ticket56

marked the checklist item The congestion window only should be updated with a congestion signal once per window as completed

marked the checklist item If the local orconn is blocked, that should be an immediate congestion signal. Also doc that we may have too large a queue there as completed

marked the checklist item If the edge connections do not have data to send, estimates of BDP should not be updated as completed

marked the checklist item I made a congestion control algorithm that directly uses the current BDP estimate as its current congestion window, and this works. We should spec it and evaluate it in Shadow. as completed

marked the checklist item BDP estimation algs as completed

marked the checklist item Westwood may have a runaway condition where max RTT continues to grow. We may want to reduce the max RTT measurement upon congestion as completed

marked the checklist item Update the consensus parameter list and tuning experiments section as completed

removed Backlog label

added Doing label

Ok Prop 324 is now up to date to match the implementation so far: !38 (merged)

closed