[e2e] Tahoe higher throuhgput than reno using finite data source

[Peter Dimopoulos]

I have been doing some simulations with Tahoe and Reno with a 
finite data source (on-off source with exponentially distributed burst and 
idle times) and I keep finding that Tahoe has higher throughput at most
loss probabilities. Has anyone else seen this behaviour, or do they have
any idea about how to explain it if it is in fact correct.

When I say throughput I don't mean goodput, I mean the throughput of
packets arriving at the bottleneck router.

Topology

On-Off
Source           Router           Destination
    O ------------ O ------------- O
     S                 R                   D

Parameters
MaxWindow:16
Buffer:100
Bottleneck BW (R->D) (pps):1250
Packet Loss Prob:0.1 (exponentially distributed)
Variable: R->D Delay 0.01-0.1 increment by 0.02 (seconds)
S-R Delay :0.02
Avg Idle Time:   7 seconds
Avg Burst Time: 7 seconds
Simulation Runtime 100000
Packet size 1460

Thanks for any help
 
----------------------------
Peter Dimopoulos
Computer Science Phd
RMIT University