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 have to continue downloads quite often and the download speed is not correct initially. I assume the algorithm keeps a list of the amount downloaded in the previous N timesteps. Each timestep gives a speed sample with the number of bytes downloaded divided by the time delta. The total speed is then averaged over all of these timesteps.
I believe the error is that in the initial time step the existing file size is not subtracted. Therefore you have a very high outlier in the list of speed samples, which slowly converges to the actual speed when new samples arrive. The error is gone once the buffer is full and the initial outlier is removed.
Solutions:
Quick Workaround: Instead of computing the mean, compute the median, which is not sensitive to a single outlier.
Bugfix: Subtract the filesize from the amount downloaded for the first timestep.
The text was updated successfully, but these errors were encountered:
When resuming a transfer, the number of bytes already transferred are taken in to account by [source:trunk/source/ch/cyberduck/ui/threading/TransferBackgroundAction.java#L146 resetting] the meter. Added test in bf4990e.
Hi,
I have to continue downloads quite often and the download speed is not correct initially. I assume the algorithm keeps a list of the amount downloaded in the previous N timesteps. Each timestep gives a speed sample with the number of bytes downloaded divided by the time delta. The total speed is then averaged over all of these timesteps.
I believe the error is that in the initial time step the existing file size is not subtracted. Therefore you have a very high outlier in the list of speed samples, which slowly converges to the actual speed when new samples arrive. The error is gone once the buffer is full and the initial outlier is removed.
Solutions:
Quick Workaround: Instead of computing the mean, compute the median, which is not sensitive to a single outlier.
Bugfix: Subtract the filesize from the amount downloaded for the first timestep.
The text was updated successfully, but these errors were encountered: