Algorithmic approach to estimate variant software latencies for latency-sensitive networking
Henning Puttnies, Björn Konieczek, Jakob Heller, Dirk Timmermann, Peter Danielis · 2016
Recently, there has been the arise of latency-sensitive IoT applications, e.g., in the field of telemedicine or Industrial Internet. Such applications have stringent latency requirements. The entire end-to-end latency between two devices is composed of several individual latencies: the software latency of the application, the software latency in the networking stack, the hardware latency on wire, and the hardware latency in switches. The hardware latencies are relatively invariant (constant). Contrary, the software latency is highly variant and thus hard to determine. However, the software-introduced latency is crucial for latency-sensitive applications, since the end-to-end latency equals the sum of all latencies between two devices. In the literature, different investigations based on simulations as well as practical test-beds using specialized hardware are proposed. In contrast, we propose a novel method using no specialized hardware that precisely estimates the latency in the networking software (precision: approx. 144μs). Our approach bases on the measurement of round-trip times between several devices and solving the resulting system of equations to estimate the latency of every individual device. Compared to the state-of-the-art the proposed method has also several advantages regarding scalability and resiliency.