Statistical Analysis of Ping data

The modified NIKHEF Ping, with the added features of Median, Interquartile Range & Percentiles, was used to obtain the data for various percentages of transfer rates between nereus.slac.stanford.edu and nocdev1-qos.es.net. The line was loaded with UDP packets using the TTCP codes.

Plots of Frequency of pings Vs Time (ms) for various TOS

The Maximum Transfer rate is approximately 9500kb/sec. The percentages are that of the maximum transfer rate.
(Click on the graphs to view a larger version.)

60% Transfer Rate
30% Transfer Rate
23% Transfer Rate
18% Transfer Rate
12% Transfer Rate
6% Transfer Rates
1% Transfer Rate

From the plots, we observe that the curves with TOS = null and TOS = 0 are almost identical for all loads. From this observation, we can infer that not setting a TOS is the same as setting the TOS = 0. Note that the loading at 1% transfer rate is far too small to have any effects on the ping.

Also, all the curves except that with TOS = 0 and TOS = null peak in-between 10 and 40 ms for all the transfer rates shown. It was not possible to locate a load range where TOS having precedence bits set between binary-001(decimal-1) and binary-111(decimal-7) would have different effects on ping.

Plots of Statistical details Vs TOS for different transfer rates

60% Transfer Rate
30% Transfer Rate
23% Transfer Rate
18% Transfer Rate
12% Transfer Rate
6% Transfer Rates
1% Transfer Rate

From the above plots, we see that the mean & median are very high for TOS =0 & null. One would expect this for significant transfer rates. But here again, there cannot be any concrete inference made to distinguish between plots having TOS with precedence bits (decimal-1 to 7). The interquartile range varies from plot to plot and thus is insufficient for making inferences.

Revised 5 August 1999
URL: http://www-iepm.slac.stanford.edu/monitoring/voip/pingstats.html
Comments to iepm-l@slac.stanford.edu