Baseline Experimental Results

1. Description:

Study the throughput of one rapid flow on different network paths in the tesbed and between different pairs of machines and compare the results with Cubic. This experiments offer us the baseline results which will help us to analyze the behaviour of Rapid for future experiments.
Machines running rapid flows: smithfd-linux1 to smithfd-linux2, smithfd-rapid1 to smithfd-rapid2 and smithfd-rapid21 to smithfd-rapid22 Rapid Version: rapid_v0.9
Network Paths: both the switched path and the routed path in netlab
TCP Configuration: All default sysctl settings. Linux autotunes buffer size and window size. MAX_IN_FLIGHT=DEFAULT=2000
RTT: default rtt on the switched path; add 0,10,30,50ms to rtt on the router path respectively

2. Experimental Results:

  • Baseline for single cubic flow from rapid1 to rapid2
  • Baseline for single cubic flow from linux1 to linux2
  • Baseline for single rapid flow from rapid1 to rapid2
  • Baseline for single rapid flow from linux1 to linux2

    The average throughput of the experiments above is (the unit is Mbps):

    flow 1 is the flow from smithfd-linux1 to smithfd-linux2. Flow 2 is from smithfd-rapid1 to smithfd-rapid2. 
                                     cubic flow1    cubic flow2    rapid flow1    rapid flow	
    no delay via switched path       940            941            865            740
    no delay via routed path         941            941            884            708
    10ms delay via routed path       933            932            532            569
    30ms delay via routed path       522            530            80.7           72.6
    50ms delay via routed path       333            345            17.9           16.8