A Matching-theoretic Framework for Consolidation of Flexible Cloud-native Central Units in 5G-RAN

In this study, we investigate the flexible many-to-one mapping of central units (CUs) to compute servers in 5G Radio Access Network (5G-RAN) using two-sided matching theory considering spatio-temporal tidal traffic patterns. Initially, we use a well known bin-packing heuristic known as First Fit Decreasing (FFD) to obtain suboptimal CU to compute server mapping. To address the traffic heterogeneity, we formulate a novel strategy of dynamic reassignment (known as Machine Admission Game or MAG) among a set of CUs and compute server in each mapping interval, using analytical approaches of coalitional game theory and college admission game. To solve this, we devise a modified version of the classical Deferred Acceptance Algorithm (DAA) satisfying the resource constraints of compute servers. We assess the benefit of the proposed matching theory framework with baseline FFD in terms of compute resource multiplexing gain (i.e., number of active servers in the CU pool) and the number of relocations incurred in the dynamic reassignment of candidate CUs. We observe that the proposed MAG framework though consumes nearly 6.8% and 4.1% of more servers than baseline FFD, it reduces the number of relocations by 33.9% on weekdays and 25.7% on weekends as compared to FFD.