Conclusion

From the comparisons in this chapter I conclude that,

• running ping in single-user mode with output to the RAM disk can reduce the source processing delay;
• whether running in multi-user mode or in single-user mode ping's own RTT measurement may be significantly different from WRTT;
• running tcpdump to capture ping traffic can improve the WRTT measurement accuracy, even when tcpdump runs on the pinging host;
• running tcpdump on a separate machine can further improve the WRTT measurement accuracy;
• the hardware configuration of a monitoring machine plays an important role in time-related measurement, for example, machines with the same hardware configurations are more likely to generate consistent tcpdump measurements than machines with different hardware configurations;
• multi-user mode does not necessarily prevent accurate time-related measurement, tcpdump can make accurate time measurements on a lightly-loaded machine;
• discrepancies in RTTs as measured by tcpdump on two identical and dedicated monitoring machines are probably due to relative clock drift between the two machines.

The most important conclusion is that in any critical time-related measurements it is crucial to check the measurement data in order to find and eliminate measurement errors.