You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I use the congestion tab quite often for investigating QoS issues. It works great for Reno/CUBIC, but it's less useful after implementing BBR. The congestion window still exists, but BBR primarily limits the send rate via a pacer.
It would be nice if we could visualize the optional pacing_rate in recovery:metrics_updated. This is measured in bits/sec so it wouldn't fit with the other graphs. It would also be nice to add a new field called bandwidth_estimate to that event, graphing it alongside the pacing_rate.
The text was updated successfully, but these errors were encountered:
It's in bits/sec so indeed wouldn't fit directly with the current y-axes, but I assume you would still like it plotted on the "top" graph, along with cwnd/loss etc.? e.g., as a separate line with its own y-axis/scaling? Or would it make more sense to have a "third" graph, in between the two current ones for example?
What would the units of the bandwidth_estimate field ideally be in your opinion?
Could you maybe provide 1 or more example qlog files with "realistic" behaviour for those parameters? So I can make sure e.g., the auto-scaling of the y-axis works well for normal values. For this, feel free to add the bandwidth_estimate field to the event yourself, using the units from question 2 :)
Since I know the FB people have also been using qlog for BBR, maybe they have some input here as well, so I can do as much as possible in 1 go? @mjoras@lnicco?
I use the congestion tab quite often for investigating QoS issues. It works great for Reno/CUBIC, but it's less useful after implementing BBR. The congestion window still exists, but BBR primarily limits the send rate via a pacer.
It would be nice if we could visualize the optional
pacing_rate
inrecovery:metrics_updated
. This is measured in bits/sec so it wouldn't fit with the other graphs. It would also be nice to add a new field calledbandwidth_estimate
to that event, graphing it alongside thepacing_rate
.The text was updated successfully, but these errors were encountered: