Curl has had the ETA algorithm problem solved for quite some time.
Roughly speaking, you keep a running average of the rate; if the rate stays constant, you remove the passed time since last-sample. Else, the ETA will be adjusted to match the running average (higher-than-average rate: ETA jumps down; lower-than-average rate: ETA jumps up).
Over a relatively short period of time, this becomes shockingly precise. Even if we only kept the running average from execution time, it would quickly become great.
HalosGhost
commented
6 years ago
Curl has had the ETA algorithm problem solved for quite some time.
Roughly speaking, you keep a running average of the rate; if the rate stays constant, you remove the passed time since last-sample. Else, the ETA will be adjusted to match the running average (higher-than-average rate: ETA jumps down; lower-than-average rate: ETA jumps up).
Over a relatively short period of time, this becomes shockingly precise. Even if we only kept the running average from execution time, it would quickly become great.