DeepSLOs for the Computing Continuum

The advent of the computing continuum, i.e., the blending of all existing computational tiers, calls for novel techniques and methods that consider its complex dynamics. This work presents the DeepSLO as a novel design paradigm to define and structure Service Level Objectives (SLOs) for distributed computing continuum systems. Hence, when multiple stakeholders are involved, the DeepSLO allows them to plan the overarching behaviors of the system. Further, the techniques employed (Bayesian networks, Markov blanket, Active inference) provide autonomy and decentralization to each SLO while the DeepSLO hierarchy remains to account for objectives dependencies. Finally, DeepSLOs are represented graphically, as well as individual SLOs enabling a human interpretation of the system performance.