On Thursday 30 September 2004 06:08 pm, Nikolaus Schulz wrote: > > and so naturally if nothing is being > > downloaded for long periods of time it'll say "stalled". I consider this > > a feature. > > If it's a feature, it should catch the traffic peaks, but it doesn't. > Probably it just doesn't poll the traffic load often enough. You have it backwards: you're asking for apt to recalculate this information *less* often (or to keep a more detailed history). What's happening is that apt notices that no data was received in the last 6 seconds, and then reports that. The formula is (amount of data received in 6 seconds)/(time since last update). This could be smoothed out by making the interval longer, but it would mean that you get a less current value. Another option is to use some sort of exponential decay (do a weighted average with the previous value), but that's more complicated and means you have to figure out some sort of threshold below which the download is considered "stalled". You could also use a moving window (keep a circular queue of the data rate seen in, say, 10 6-minute slices) and take the average of those samples (weighted by the legnth of the slice, of course). This method has the serious drawback that it would change the apt ABI, since you'd have to add the queue to the pkgAcquireStatus object (thus changing its size). And in the end, any of these methods will still miss "bursty" transmissions of the type you describe; all they do is increase the interval within which data must arrive. ie, no matter how big you make the window, someone can still have a connection that doesn't send any data within the window (and it sounds like apt-proxy would trigger any interval, if it downloads entire files before streaming them to clients). The only reasonable thing that can change (IMO) is to increase the interval over which apt calculates data rates and maybe to use an exponential-decay technique, and you can negotiate with the apt maintainers about that. (personally, I think that simply increasing the interval will make the progress indicator too insensitive to variations in the transfer rate) Daniel -- /------------------- Daniel Burrows <dburrows@debian.org> ------------------\ | "Fluble, the others want you to know that we | | have you surrounded with tranquilizer rifles | | and are prepared to use them. Again." -- Fluble | \---------------- The Turtle Moves! -- http://www.lspace.org ---------------/
Attachment:
pgpb9hElWXUaS.pgp
Description: PGP signature