Experimental Results for Rapid Transport Protocol

Testbed tools:

Using rapid_v0.9

1. Baseline Experiments

-- throughput of one Rapid flow on different network paths and with different round trip times

2. Effects of TCP Buffer and Window Size

-- Study how the congestion window size and tcp buffer size affect the throughput of one rapid flow with different round trip times

3. Rapid with UDP Cross Traffic

-- Study how one rapid flow reacts to ON/OFF UDP Cross Traffic

4. Intra-protocol Fairness

-- Study the throughput of two coexisting rapid flows

5. Over 10G Link on ANI Testbed

-- The behaviour of one rapid flow over 10Gbps link

6. Summery of Problems and Suggestion for Modification

Using rapid_v0.9.2

1. How iperf udp perform as cbr cross traffic:

2. Correcting the "packet rejects because of timestamp" problem:

3. Experiments with UDP cross traffic:

4. Do we need scheduled send gaps smaller than what the NIC card can send?

5. Turn off Interrupt Coalescence:

-- Problem narrow down to AB_est calculation

6. Futher study of Rapid with 300M Cross Traffic:

-- UDP cross traffic is not strict CBR

6. Try Averaging Mechanism to Get Proper AB_est:

Continued:
Exponential Moving Averaging:
Comparison Between two averaging schemes:

7. One Rapid Flow with Thttp Cross Traffic

Updated pruning algorithm: new_pruning_algorithm.txt

8. Re-analysis for 300Mbps CBR cross traffic:

9. Different Pstream Size -- one rapid flow with two thttp flows


pstream size = 80

pstream size= 720

Using Rapid_v0.9.4

Experiment with one rapid flow alone

more experiments:

One Rapid Flow with 500Mbps UDP cross traffic:

Comparison Between pstream length, smoothing algorithm for ABests:

Spike Removing ALgorithm with different CBR:

Study of SendRate:

Analysis of one rapid flow with thttp traffic with different parameters:

Using Rapid_v0.9.4_spike

With UDP ON/OFF CBR:

A New ABestimation Proposal:

360-packet probe stream:

Repeated Experiments:

What if logging on the receiver side?

Probe stream with 360 packets each with thttp cross traffic

Probe stream with 80 packets each with thttp cross traffic

Thttp Baseline

Using Rapid_v0.9.5

experiments with different paramenters

experiments with selected parameters and Cubic/Reno

Varying TAU/ETA for 160/10/0

What parameters give best abests? => ABestimations of 300Mbps UDP:

Rapid with Multiple thttp flows emulating rtt:

On ANI-10G testbed: