Quoting Christian Biere <[EMAIL PROTECTED]> from ml.softs.gtk-gnutella.devel: :I think the default chunk sizes (min/max: 512 KiB, 10 MiB) are far :too big especially considering swarming. I'd like to reduce these :to 64 KiB (min) and 256 or 512 KiB (max). At a later time this could :be adjusted depending on the available bandwidth.
I'm not sure that would be a good move. :Anyway, I even suspect that these huge chunk limits are responsible :for lossage of downloaded data at shutdown time. E.g., consider :you're currently downloading from 4 sources 10 MiB each and shut :down shortly before these are finished. The chunks are marked as :busy and will thus be reset to empty on startup causing lossage :of up to 40 MiB - if my assumption is correct, since I didn't :verify this at the moment. No, no. That cannot be unless there's a bug: the lower boundary of each "busy" chunk is marked as "done" whenever data is received. And the fileinfo is synced on a regular period, so in the advent of a crash, you loose the amount of data downloaded since the update, but not the whole downloaded part of the chunk! :Smaller chunks would also decrease the turn-around time for :swarming so that peers could sooner contribute. If you look at :your upload screen, you'll probably see that all other vendors :use much smaller chunk sizes (dunno this might depend on your :connection speed if they handle this dynamically). Not very :familiar with PARQ, this also seem to cause Gtk-Gnutella peers :getting pushed far down in the queue due requesting up to 10 MiB. Again, since the lowest boundary of the chunk is updated as soon as data is available, even if you request 10 MiB for instance and got the first 512 KiB of it, you can swarm those 512 KiB. :This means also that Gtk-Gnutella blocks upload slots for :much longer than other peers resp. allow them to block slots :much longer. Some peers will request 100 chunks of 100 KiB and GTKG will request one chunk of 10 MiB. The slot will be actually blocked much longer by other peers because of the latency between requests. I'd leave the current range of the chunk size as it is. Raphael ------------------------------------------------------- This SF.Net email is sponsored by Oracle Space Sweepstakes Want to be the first software developer in space? Enter now for the Oracle Space Sweepstakes! http://ads.osdn.com/?ad_ids93&alloc_id281&op=click _______________________________________________ Gtk-gnutella-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/gtk-gnutella-devel
