End-to-end (E2E) reliability provisioning is critical for network slice customization. Leveraging martingale theory, we construct an analysis framework of the reliability with regard to E2E latency for the multihop system, where service processes provided by wireless nodes and networking nodes are heterogeneous. Based on the Wald martingale constructions of arrival processes and service processes, tandem service descriptor is defined to embody the features of tandem service mode for the targeted flow. Arrival bias and service bias are derived to describe the fluctuations of arrival and service processes, respectively. Relying on the Doob maximum inequality of martingales, a tight upper bound of the unreliability regard to E2E latency is captured. Packet duplication and hot-backup parallel transmission patterns are adopted for network reliability enhancement. A reliability decomposition and bandwidth abstraction algorithm is proposed, which achieves the decoupling between the statistical reliability requirement and the desired bandwidth of each node. Slice instantiation is performed in the access network and the core network, respectively. In the wireless access network, a long-term stochastic optimization problem is formulated with reliability requirements constraint. Lyapunov drift-plus-penalty (DPP) optimization theory is explored to transform the intractable long-term optimization problem into dynamic evolutionary per-slot optimization problems. We yield the closed-form solutions of user scheduling and power allocation. In the core network, the forwarding rates in the networking nodes for the targeted flow are obtained.

Based on martingale theory, the authors explore a novel method to analyse the bound of delay violation probability for aggregate traffic in this study. The proposed arrival model, four‐state Markov modulated multinomial process, describes the behaviour of mixed service flows, which depicts the packet arrival process of two interweaved services and highlights the arrival characteristics of the batch, bursty, and heterogeneous. The traffic of data service is modelled by the interrupted multinomial process to depict the randomness and discontinuity. Relying on martingale theory, arrival‐supermartingales provide a model for integrating multiple heterogeneous services. They construct a supermartingale process of the backlog in the buffer, which enables the analysis of delay for the queuing system with multiple arrivals and a single server. Finally, the delay of aggregate traffic is featured by the parameters of its corresponding supermartingales. Based on the stopping time theory of martingales, the delay violation probability bound is derived, which matches the simulation results well and is tight.