Averaging Among Pruned Segments When Calculating AB_est for Each Probe Stream

Description: Caculate sendgap and receive gaps for multiple acknowledged segments in one probe stream when getting AB_est to remove the burstiness of receive gaps. To get samples as many as possible, the averaging mechanisim is done in the way of sliding-window. And meanwhile try different GAP_NS_EPSILON heuristic.

Throughput of UDP Cross Traffic for Each Pstream



No Averaging Mechanisim




Gap of AB= 35663
probe_id= 10152 pstream_abest= 924.973688167193 prune_size= 39 pstream_size= 0 excursion= 1 excursion_from_beginning= 0
probe_id= 10153 pstream_abest= 955.065105748567 prune_size= 37 pstream_size= 0 excursion= 1 excursion_from_beginning= 0


Averaging Among Every Two Acked Segments




Gap of AB= 71326
probe_id= 10152 pstream_abest= 832.546730948153 prune_size= 38 pstream_size= 0 excursion= 1 excursion_from_beginning= 0
probe_id= 10153 pstream_abest= 838.03296553603 prune_size= 36 pstream_size= 0 excursion= 1 excursion_from_beginning= 0



Averaging Among Every Three Acked Segments





Gap of AB= 106989
probe_id= 10152 pstream_abest= 708.589001943192 prune_size= 37 pstream_size= 0 excursion= 2 excursion_from_beginning= 0sudden decrease 
probe_id= 10153 pstream_abest= 689.42846469929 prune_size= 35 pstream_size= 0 excursion= 1 excursion_from_beginning= 0sudden decrease 



For some probe streams that over-estimate the available bandwdith:





Averaging Among Every Four Acked Segments





Gap of AB= 142652
probe_id= 10152 pstream_abest= 713.337391657246 prune_size= 36 pstream_size= 0 excursion= 1 excursion_from_beginning= 0
probe_id= 10153 pstream_abest= 669.2867342082 prune_size= 34 pstream_size= 0 excursion= 1 excursion_from_beginning= 0sudden decrease 



CDF of AB_ests for Probe Streams When There's 300Mbps UDP Cross Traffic