Can Reliability be a consistent metric in radio access networks? A scheduler based illustration

Mobile networks of the future (5G and beyond) are conceived to facilitate technologies that pervade daily human life and cater to every one of the numerous services these technologies solicit. Faced with a staggering diversity in service requirements, researchers or engineers often turn to reliability as a means to appraise the efficacy of networks. What complicates the matter, however, is the fact that 'reliability' is a word one often encounters in common parlance. This, therefore, results in works which end up dealing with a definition of reliability in a colloquial linguistic sense rather than as a measure (or metric), as defined in mathematics. This work - by defining reliability as a probability measure - provides a quantitative definition of reliability, which stems from its characterization in the dictionary and is based on quantifiable definitions of resilience, availability, and other parameters important to radio access networks. The practicality and consistency of this metric are then illustrated by designing a reliability-aware scheduler, which exploits predictions of the users' channel quality and directly incorporates the metric introduced in order to allocate resources to various concurrent services. The optimal scheduler is proven to be NP-hard and a heuristic is proposed as its consequence. Numerical evaluations show that the reliability-aware scheduler and its heuristic outperform the well-known proportional fair scheduler, not just when the predictions are error-free, but also when the predictions of the users' channel gains are error-prone.