On Thu, 14 Oct 1999, Chris Frey wrote: > Does CurrentBytes not get updated with all bytes coming in on a > parallel download? I thought it did.
It does, but on non-modems where bandwidth is limited by the interconnect and not the local connection you get a sharp drop off when one parallel site runs out of files. So the cps might go from 200k/s -> 100k/s which obviosly is messed up by your patch. > I also don't see how a large inter-file delay could cause a problem. > Such a delay would add to the total time the download is taking. No, a delay like that adds to the total time it has taken, not to the amount of time it does take and it should not be included in an eta calculation. Consider that you downloaded at a steady 1.5k/s for 10s then paused for 3s to get a new file then downloaded the remainder at 1.5k/s for another 10s. This yeilds an average ETA of 1.2k/s at time 15s and would give an ETA of 10 seconds - 2 seconds off the real eta of 8 seconds. As time goes on the average rate will slowly increase, but the ETA will fall -faster- than 1 eta second per real second so although it appears to be accurate it is 100% bogus. Instanenous ETA gives you a new ETA base every 5 seconds and you can choose on your own which is most likely. It works amazingly well on lan connections but modems that randomly drop stuff it sucks just as much as everything else out there. Jason
