Simulation Setup S C1C1 CNCN T_Sink1 T_SinkN T1T1 TNTN U U_Sink
4N Mbps 50 ms L Types of traffic: S -> C i {1..N} : Joint-TCP
flows T i {1..N} ->T_Sink i {1..N} : Ordinary TCP Reno flows U
-> U_Sink : Background Poison UDP traffic at the rate of 40x2xN
kbps Others: TCP MSS: 1460bytes TCP advertised window: 64Kbytes UDP
Pkt Size: 500 bytes Queuing Decipline: DropTail N : # of Joint-TCP
flows Links: All links except link L have bandwidth of 100Mbps and
propagation delay of 10ms 0.98) regardless of the number of
Joint-TCP streams and the allocation ratio The performance
difference when RED is adopted is negligible.">
Allocation Accuracy (Cont) Summary: The allocation accuracy of
MulTCP (with or without SACK) degrades when the allocation ratio
increases (e.g. from 1:2:3 to 1:4:8). Buddy-TCP maintains high
allocation accuracy (>0.98) regardless of the number of
Joint-TCP streams and the allocation ratio The performance
difference when RED is adopted is negligible.
Slide 8
Impact Ratio Impact Ratio for Droptail
Slide 9
Impact Ratio (Cont) Impact Ratio for RED
Slide 10
Impact Ratio (Cont) Summary: MulTCP (with and without SACK)
causes an decrease in the average throughput of the competing
ordinary TCP traffic. The decrease can be upto 70% when SACK is
used. Buddy-TCP causes an increase (10 20%) in the average
throughput of the competing ordinary TCP traffic. RED has
negligible effect on the Impact Ratio
Slide 11
Effect of Random Packet Loss S C1C1 CNCN T_Sink1 T_SinkN T1T1
TNTN U U_Sink 4N Mbps 50 ms L -Random packet loss is introduced in
the bottleneck link L -N = 3
Slide 12
Effect of Random Packet Loss (Cont) Allocation Accuracy for the
allocation ratio (1:2:3)
Slide 13
Effect of Random Packet Loss (Cont) Allocation Accuracy for the
allocation ratio (1:4:8)
Slide 14
Effect of Random Packet Loss (Cont) Impact Ratio for different
loss probability
Slide 15
Effect of Random Packet Loss (Cont) Summary When random packet
loss probability is increased at the bottleneck link, the
allocation accuracy will be increased, and the impact ratio will
approach to 1. The effect is more apparent for MulTCP (with and
without SACK) Reason: When random packet loss probability increase
to certain level, the total throughput is NOT limited by the
bottleneck bandwidth anymore. Probability of TCP Timeout due to
burst of packet losses at the bottleneck node is reduced.
Slide 16
Whats Next? Convergence of bandwidth allocation Vs RTT
Application on adaptive video streaming (some preliminary
results)