Simulations for "Dynamic Behavior of Slowly-Responsive Congestion Control Algorithms"

postscript[.ps], pdf[.pdf]

by Deepak Bansal, Hari Balakrishnan, Sally Floyd and Scott Shenker
Sigcomm 2001, August 27-31, San Diego, CA.

This page has pointers to the simulation scripts for the above paper. These simulation scripts are also available from a tar zipped file . The README file describes the procedures for running these simulations.

These scripts use the latest snapshot of ns (ns-2.1b7a-current as of June 20,2001). Since randomization is enabled in all of these simulations, the precise results of simulations can be different from those in the paper and can also change with different versions of NS.

The simulation for Figure 3 on "the droprate for several slowcc algorithms" can be run with:
droprate.tcl, run-RED.tcl and plot-RED (for generating postscript file)..

Figure 4 and 5 on "the stabilization time" and "the stabilization cost" for various slowCC flows can be obtained using:
dropratemetric.tcl, run-RED.tcl , avg and plot-RED.

(run-RED.tcl produces the result in a file result-RED.  The average values are generated using avg script.You need to then, split that avg-result-RED file generated as a result, manually into 5 files (corresponding to each slowCC algorithm) , avg-result-RED-1 through avg-result-RED-5 and replace 1-1/b values with b. (eg. 0.5 with 2, 0.875 with 8, 0.995 with 256 etc.). plot-RED uses these files to generate the result. )

The simulations for Figure 6 on "flash crowd simulations" (top, middle, bottom) can be run with droprate.tcl, flash1000run.tcl and flash1000plot.

These simulations take quite long to run (I usually leave them in night and they are complete by morning; 1000 short flows really take very long to run). Sometimes, RAP with b = 0.5 crashes. I suggest, in that case, rather than rerunning the simulations all over again, run the remaining simulations by modifying flash1000run.tcl. I do not know the reason for RAP crashing. Also, the flash1000run.tcl generates results for many SlowCC algorithms and you may want to cut that file to run only the simulations for which results are given in the paper (i.e. TC(1/2), TFRC(256. no self clocking) and TFRC(256,self clocking)).

Figures 7,8 and 9 (oscillating bandwidth simulations with long-lived competing SlowCC flows) can be obtained by using the following files:
osc-bw.tcl, percent1, avg1, run1.tcl and plot1.

The simulations for figures 10 can be run with droprate.tcl , run.tcl and avg . Figure 11 is not generated by simulation (hence no scripts) but is the mathematical function derived in the paper for delta-fair convergence times, and has been plotted using gnuplot.

The simulations for figure 12 can be run with droprate.tcl, run.tcl, plot and avg.

The simulations for figures 14 and 15 (oscillating bandwidth simulations with identical flows) can be run with osc-bw.tcl, percent, avg, run.tcl and plot.

Simulation scripts for figures 17,18 and 19 are available here .

Last modified: June 2001 by Deepak.