On Wed, Apr 12, 2006 at 12:49:16PM -0700, Ian Clarke wrote: > > On 12 Apr 2006, at 12:21, Michael Rogers wrote: > > >>Increment the window size by 1/window size when a packet succeeds. > >>Halve it when one fails. > > > >That's what TCP does in congestion avoidance mode (ie when cwind > > >ssthresh) but it's less aggressive than what TCP does in slow start > >mode, so connections will take longer to reach full speed and to > >return > >to full speed after a packet loss. You might not get great link > >utilisation but it seems like a good compromise because it avoids the > >complexity of calculating ssthresh. > > If we don't already, we should also remember the window size between > the transmission of different blocks of data, rather than starting > from scratch once every 32kb block.
We should have one limiter between each pair of peers, rather than trying to throttle it for each block, IMHO. > > Ian. -- Matthew J Toseland - toad at amphibian.dyndns.org Freenet Project Official Codemonkey - http://freenetproject.org/ ICTHUS - Nothing is impossible. Our Boss says so. -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 189 bytes Desc: Digital signature URL: <https://emu.freenetproject.org/pipermail/devl/attachments/20060412/73a4bbda/attachment.pgp>
