Delay Measurement Methodology Revisited: Time-Slotted Randomness Cancellation

Abstract

The operation of today's networked applications and protocols depends to a large extent on accurate end-to-end delay measurements and estimations. Appropriate methodologies can significantly improve quality, accuracy, and representativeness of delay samples acquired through measurements. This paper focuses on limitations of state-of-the-art end-to-end delay measurement methodologies. Its central observation is that the first time-slotted link in a measurement path cancels start-time randomness of delay measurement samples. When leaving this first link, all measurement samples are time-synchronized with each other and potentially with global time modulo link period. Because of this effect, which the paper introduces as time-slotted randomness cancellation effect, random sampling is not possible beyond the first time-slotted link of measurement paths. End-to-end measurements, therefore, fail to capture the full delay range of subsequent time-slotted links in the path, measurement results being limited to a specific session setup. Following a detailed discussion of theoretical models, the paper proposes a novel delay measurement methodology that adds artificial delay functionality to intermediate network nodes. Measurement packet headers include random seeds that are used by compatible ingress nodes of subsequent time-slotted network segments to regenerate start-time randomness. Samples acquired with this measurement methodology can assess a network path's full delay range. Practical applicability of the presented concept and methodology is demonstrated by a prototype implementation that assesses delay in public mobile cellular networks. Measurement results presented in this paper confirm that the proposed concept and methodology is generally applicable and that randomness regeneration can significantly improve quality and representativeness of delay measurement samples.