Simulation Suite for Comparative Studies of PS, SRPT, ViFi and SFS

This page contains links to the simulation code and instructions for generating the results reported by Sergey Gorinsky and Christoph Jechlitschek in the ICCCN 2007 (Best Paper Award Nominee) article "Fair Efficiency, or Low Average Delay without Starvation". The results were originally produced as per the given instructions under Cygwin.

• Figure 3a, "Fairness of ViFi and SFS versus unfairness of SRPT for uniformly distributed message sizes at 95% load: starvation stretch for every seventh message in a single experiment"
  
  
  1. Compile:
           gcc -o figure3a.out -lm figure3a.c
     
  2. Execute:
           ./figure3a.pl
     
  3. Plot the graphs:
           gnuplot < figure3a1.gnu
           gnuplot < figure3a2.gnu
     
  4. View the graphs:
           gv figure3a1_bw.ps&
           gv figure3a2_bw.ps&
     
  
  
• Figure 3b, "Fairness of ViFi and SFS versus unfairness of SRPT for uniformly distributed message sizes at 95% load: average of starvation stretches over 1,000 experiments; the rich get richer under ViFi or SFS, and so do the poor; SFS helps the middle class more"
  
  
  1. Compile:
           gcc -o figure3b.out -lm figure3b.c
     
  2. Execute:
           ./figure3b.pl
     
  3. Plot the graphs:
           gnuplot < figure3b1.gnu
           gnuplot < figure3b2.gnu
     
  4. View the graphs:
           gv figure3b1_bw.ps&
           gv figure3b2_bw.ps&
     
  
  
• Figure 3c, "Fairness of ViFi and SFS versus unfairness of SRPT for uniformly distributed message sizes at 95% load: cumulative distribution of individual starvation stretches from 100 experiments"
  
  
  1. Compile:
           gcc -o figure3c.out -lm figure3c.c
     
  2. Execute:
           ./figure3c.pl
     
  3. Plot the graphs:
           gnuplot < figure3c1.gnu
           gnuplot < figure3c2.gnu
     
  4. View the graphs:
           gv figure3c1_bw.ps&
           gv figure3c2_bw.ps&
     
  
  
• Figure 4, "Efficiency of SRPT, PS, ViFi, and SFS with uniformly distributed message sizes."
  
  
  1. Compile:
           gcc -o figure4.out -lm figure4.c
     
  2. Execute:
           ./figure4.pl
     
  3. Plot the graphs:
           gnuplot < figure4a1.gnu
           gnuplot < figure4a2.gnu
           gnuplot < figure4b1.gnu
           gnuplot < figure4b2.gnu
           gnuplot < figure4c1.gnu
           gnuplot < figure4c2.gnu
     
  4. View the graphs:
           gv figure4a1_bw.ps&
           gv figure4a2_bw.ps&
           gv figure4b1_bw.ps&
           gv figure4b2_bw.ps&
           gv figure4c1_bw.ps&
           gv figure4c2_bw.ps&
     
  
  
• Figure 5, "Efficiency of SRPT, PS, ViFi, and SFS with Pareto-sized messages"
  
  
  1. Compile:
           gcc -o figure5.out -lm figure5.c
     
  2. Execute:
           ./figure5.pl
     
  3. Plot the graphs:
           gnuplot < figure5a.gnu
           gnuplot < figure5b.gnu
           gnuplot < figure5c.gnu
     
  4. View the graphs:
           gv figure5a_bw.ps&
           gv figure5b_bw.ps&
           gv figure5c_bw.ps&
     
  
  

Christoph Jechlitschek and Sergey Gorinsky
chrisj@arl.wustl.edu and gorinsky@arl.wustl.edu