Problem: To investigate the use of the ping/spray UNIX commands to accurately predict delays in an Internet environment.

Experiment: Initially, a 2k-Factorial experimental design was used to collect data. The following factors and levels were used as a measure:

• Location
  • University of Maryland (umd)
  • UCLA
• Day
  • Weekday
  • Weekend
• Time of day
  • Peak
  • Off-Peak
• Day
  • Weekday
  • Weekend
• Size of Ping packet
  • 68
  • 1008
• Number of Pings
  • 5
  • 50

 To aid in collecting the large amount of data needed for this experiment, a UNIX script file was written (see Appendix A). After running the script file for a week, the data was entered into Table 1 (each number is an average of the table entry). A result table (Table 2) was then constructed to calculate the main effects of varying each factor on the response variables (MEF) as well as the average interaction effect of each factor (IEF). The MEF and IEF results indicate that varying the location and packet size factors affects the outcome of the experiment, whereas varying the three other factors has almost no effect at all and, therefore, are negligible.

  With this in mind, repeating the experiment, collecting more data to perform a Full Factorial Design experiment. The following factors and levels used in this design are:

• Location
  • University of Maryland (umd)
  • University of Wisconsin
  • Indiana University
  • University of Texas
  • UCLA
• Day
  • Weekday
• Time of day
  • Peak
• Day
  • Weekday
• Size of Ping packet
  • 68
  • 303
  • 538
  • 773
  • 1008
• Number of Pings
  • 5

 The script file was then adjusted for collecting the data (see Appendix B). The results were averaged each day and inputted into Table 3. From this, it was possible to construct two linear regression models, one for the ping command, and the other for the spray command. The confidence intervals were also computed from the results of Table 3.

 At this point, a constant was included with the ping regression equation giving the following equation:

Y = 1425.99319149 + (C + 20.6412765957) * X

where X is the value of each spray packet size.

 The slopes of the ping regression equation, with constant and spray regression equation, were set equal to each other, and the resulting constant was found to be: 13.116.

 Calculation: Spray regression model: y = 563.2046 + 7.525X

At this point, a new spray script file was written to collect the average values of the spray (bytes/second) for the various packet sizes (see Appendix C). The data was, again, plotted to Table 5.

Conclusion: It can be seen from Table 5, the ping regression line doesn’t fall within the confidence intervals of the spray data collected. Therefore, the stated hypothesize is that this experiment shows that the UNIX ping command cannot predict the outcome of the spray command.