#FIG 3.1 Landscape Center Inches 1200 2 2 4 0 6 -1 7 0 0 -1 0.000 0 0 7 0 0 5 12300 9450 12300 150 225 150 225 9450 12300 9450 2 2 0 6 -1 7 0 0 -1 0.000 0 0 -1 0 0 5 1050 2100 2625 2100 2625 3300 1050 3300 1050 2100 2 2 0 6 -1 7 0 0 -1 0.000 0 0 -1 0 0 5 6975 2100 8550 2100 8550 3300 6975 3300 6975 2100 2 2 0 6 -1 7 0 0 -1 0.000 0 0 -1 0 0 5 4050 2100 5625 2100 5625 3300 4050 3300 4050 2100 2 1 1 6 -1 7 0 0 -1 14.000 0 0 -1 1 0 2 1 0 6.00 360.00 720.00 2625 2700 3975 2700 2 1 1 6 -1 7 0 0 -1 14.000 0 0 -1 1 0 2 1 0 6.00 360.00 720.00 5700 2700 6975 2700 2 5 0 1 -1 -1 0 0 -1 0.000 0 0 -1 0 0 5 0 1425 900 10350 900 10350 3450 1425 3450 1425 900 4 0 -1 0 0 16 20 0.0000 4 255 4110 7950 9225 BRIMS:Promise of the Internet\001 4 0 -1 0 0 16 20 0.0000 4 315 1830 525 9300 May 29, 1997\001 4 0 -1 0 0 16 18 0.0000 4 210 5850 975 825 The Classic Sender-Channel-Receiver Picture:\001 4 0 -1 0 0 16 18 0.0000 4 210 900 1350 2850 Source\001 4 0 -1 0 0 16 18 0.0000 4 210 1005 4275 2850 Channel\001 4 0 -1 0 0 16 18 0.0000 4 210 540 7350 2850 Sink\001 4 0 -1 0 0 16 18 0.0000 4 270 10995 975 4275 1. Remove Artifice of periodicity (packet boundary) in source - to permit re-packetization\001 4 0 -1 0 0 16 18 0.0000 4 210 2940 975 4605 in intermediate nodes - \001 4 0 -1 0 0 16 18 0.0000 4 270 11175 975 4935 At one level, this is basically FEC or Scaalble Coding with packet marking (ghanbari, garrett\001 4 0 -1 0 0 16 18 0.0000 4 270 7755 975 5265 etc), but we could go further and make queue more like CDMA!\001 4 0 -1 0 0 16 18 0.0000 4 270 11040 975 5925 2. Remove artifice of periodicity in grabber (the frame grab model of source is wrong - we\001 4 0 -1 0 0 16 18 0.0000 4 270 9675 975 6255 want object/delta + conditional replenishment model, to reduce the variation....)\001 4 0 -1 0 0 16 18 0.0000 4 270 7020 975 6915 3. Remove size of step function on re-route for flows by \001 4 0 -1 0 0 16 18 0.0000 4 270 10935 975 7575 i) re-route all flows that are not changing rapidly (easy for marked packets in flow....if the \001 4 0 -1 0 0 16 18 0.0000 4 270 10800 975 7905 marking is not binary, but is progressive - i.e. the more change i nsource, the higher the \001 4 0 -1 0 0 16 18 0.0000 4 270 3135 975 8235 don't re-route value is....)\001 4 0 -1 0 0 16 18 0.0000 4 270 7725 975 8565 ii) constrain re-routes to not alter delay too much - also easy if\001 4 0 -1 0 0 16 18 0.0000 4 270 9510 975 8895 a) queues are short and b) only care about e2e (propagation) delay difference\001