Closed Bug 2021867 Opened 6 months ago Closed 6 months ago

Add additional slow start telemetry with new neqo v0.23.1 stats

Categories

(Core :: Networking, enhancement, P2)

enhancement

Tracking

()

RESOLVED FIXED
150 Branch
Tracking Status
firefox150 --- fixed

People

(Reporter: omansfeld, Assigned: omansfeld)

References

Details

(Whiteboard: [necko-triaged])

Attachments

(1 file)

Following plan:

  1. slow_start_exit_accuracy: A labeled_custom_distribution with labels heuristic_exit and loss_exit, ranging from 0% to 150% with a linear scale. Gives the magnitude of how far off the slow start exit was from where the connection ended up, as a percentage of the final cwnd. final_cwnd is the denominator because it is the reference/ground truth. Separating by exit reason is important because loss and heuristic exits are fundamentally different cases that would obscure each other if mixed. For now we will only record for the loss_exit label as a first heuristic is still wip in Bug 2006690, but I would like to already see the metric working and gather data for loss based slow start.
  2. slow_start_exit_direction_{loss|heuristic}: A labeled_counter with labels overshoot and undershoot. Two separate metrics, one for loss exits and one for heuristic exits. The loss one can already be implemented now. For loss exits, this primarily confirms the expected assumption that loss exits almost always overshoot. For heuristic exits, this is where the interesting signal will live, showing whether the algorithm in use tends to over, or undershoot.
  3. congestion_event_count: A custom_distribution ranging from 0 to 100. Shows how many congestion events a connection has, with spurious congestion events filtered out. Is interesting to get an intuition to how firefox connections look like in regard to congestion control. Can also be used to reason about state of other metrics, i.e. if 1. looks to be performing very well for loss_exit but most of our connection only see 1 congestion event then that means that the comparison to final_cwnd is not accurate for loss_exit because we didn't have a chance to adjust the overshoot.
Depends on: 2020969
Attachment #9551142 - Attachment description: WIP: Bug 2021867 - Add additional slow start telemetry → Bug 2021867 - Add additional slow start telemetry
Status: ASSIGNED → RESOLVED
Closed: 6 months ago
Resolution: --- → FIXED
Target Milestone: --- → 150 Branch
QA Whiteboard: [qa-triage-done-c151/b150]
See Also: → 2064979
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: