Measurement considerations for assessing unidirectional latencies
This paper presents a study of single direction latencies to selected destinations of the Internet utilizing a variety of paths. The objective is to demonstrate that round-trip latencies are an insufficient and sometimes misleading method to determine unidirectional delays. This claim has significant implications for high-speed, multi-application, wide-area, traffic aggregating networking environments which often require predictability of precise delay.
Assuming that a round trip path is symetric and dividing a RTT in half to find the latency in either the outgoing or incoming tranmission path is highly inaccurate as there are both static and du\ynamic reasons for the paths traversed by packets between hosts to be asymetric. The static reason is that the route may be permanently configured to be asymetric because of network harware limitations (for example simplex links). The dynamic reasons for asymetrically datagram paths are resource contentions between other components in the network. Digitial continuous media networked applications require good predicatability of unidirectional network delays, which simply halving the RTT cannot give.