Improving Delay Bounds in the Stochastic Network Calculus by Using less Stochastic Inequalities

Stochastic network calculus is a versatile framework to derive probabilistic end-to-end delay bounds. Its popular subbranch using moment-generating function bounds allows for accurate bounds under the assumption of independence. However, in the dependent flow case, standard techniques typically invoke Hölder's inequality, which in many cases leads to loose bounds. Furthermore, optimization of the Hölder parameters is computationally expensive. In this work, we show that two simple, yet effective techniques related to the deterministic network calculus are able to improve the delay analysis in many scenarios, while at the same time enabling a considerably faster computation. Specifically, in a thorough numerical evaluation of two case studies, we show that using the proposed techniques: 1. we can improve the stochastic delay bounds often considerably and sometimes even obtain a bound where the standard technique provides no finite bound; 2. computation times are decreased by about two orders of magnitude.