One Edge to Serve Both: A System Model for Joint 6G User Plane and Service Function Allocation
Abstract
In line with the growing cloudification, mobile networks have been increasingly moving their Network Functions (NFs) to cloud environments so they can unlock resource efficiency gains by shared use of the underlying physical infrastructure. Despite this potential, sharing resources among user plane NFs and Service Functions (SFs) may lead to unexpected performance degradation due to contention for the compute and networking resources, especially under variable load scenarios. The system models from current literature on NF placement assume deterministic, load-independent performance with strict resource partitioning and allocation. This simplification makes the models less accurate for flow level Service Level Agreement (SLA) evaluations because they do not account for traffic variability or resource load impact. Furthermore, the enforcement of strict resource partitioning and allocation is complex and impractical at scale. To mitigate these issues, we (i) propose removing the load-independence assumptions by using a scalable, queueing-based stochastic system model that accurately captures load-dependent performance and workload collocation effects in edge infrastructures, (ii) develop latency quantile approximations at the compute node level, and (iii) develop end-to-end latency quantile approximations across the core network. We validate the model and analyze its strengths and limitations in predicting network flow performance and guiding SLA-compliant workload allocation. Our analysis shows that the model predicts the endto-end core network latency with a Mean Relative Error (MRE) of 0.1 and results in less than 1% of false positives and 13% of false negatives for flow level SLA constraint evaluation.