Time remaining for downloads is unreliable

[markrian]

It seems that transmission calculates the time remaining to download a file by simply dividing the amount of data left by the current download speed. Obviously, speeds vary over the course of a download, and so then does the estimated time remaining. This can vary by quite a lot, making the estimation unreliable. It would make sense to uses some sort of smoothing, or a rolling average to give a better estimate.

KTorrent offers various methods for calculating time remaining estimates, however, their 'KTorrent' method is a combination of various methods that was found to be the most accurate in their testing. I've found it to be surprisingly accurate and reliable.

Perhaps their method could be brought to Transmission without too much trouble?

[charles]

The `KTorrent' method described in this ticket can be found at:

 http://websvn.kde.org/trunk/extragear/network/ktorrent/libbtcore/torrent/timeestimator.cpp?view=markup

[MSN_Exploder]

Hi, i wrote a small patch for using the KTorrent style ETA in Transmission.

HtH

MSN_Exploder

[charles]

MSN_Exploder: it looks like that patch uses static variables for the samples... wouldn't that mean that different torrents would get each others' samples mixed in with their own?

[MSN_Exploder]

Sorry for my late reply, kinda busy right now.

i think adding all samples into the torrent struct should help. But i'm really not into the whole C Coding... Maybe you can point me into the right direction.

[MSN_Exploder]

I revised the patch, removed all wrong statics... It should work correctly now.

(had to add a tiny ugly "hack", maybe you know a better solution)

[fulgas]

Hi, i'm sorry if i'm stepping into someone shoes but i did a almost equal eta to KT as a exercise. It took a couple of hours. I tried to follow tr code guidelines.

regards,

Nelson Silva

[geertjan]

Why is this marked as "patch needed"? There seem to be two patches available here.

Just asking, because I love Transmission but the ETA is currently quite useless, jumping from very high to very low all the time.

[charles]

There seem to be two patches available here.

Have you tried them? Give them a spin and tell me you think they're ready to go in.

[charles]

I used a simplified version of Pan's code. It was easy to drop in, seems to work reasonably well, and was implemented in 5 lines instead of 300.

Added in trunk in r8770 for 1.73.

[charles]

oops, make that r8771 :)

[geertjan]

Charles, just wanted to say thanks. I was actually planning to try the posted patches today, I didn't want to be one of those people who just ask things from open source developers without contributing anything. But I'll just look forward to version 1.73 now :)