Performance Modeling for Cloud-native Network Functions Management

This technical report presents an analysis of performance modeling for Cloud-native Network Functions (CNFs) in closed-loop management systems. The document outlines the complexities involved in allocating resources and optimizing system parameters for CNF deployments, particularly in shared resource environments. It experimentally investigates the impact of CNFs sharing CPU and memory resources, highlighting that an increase in the number of deployed CNFs leads to greater variability in packet service times, resulting in longer waiting times. The report proposes a model for CNF performance when sharing resources and discusses its scalability for runtime orchestration. Key contributions include a method for parameter estimation using Queueing Theory, and the establishment of a general compute-network resource model that captures the sharing dynamics of CPU and memory resources. The structure of the report includes sections on related work, system description, experimental setup, and model evaluation, culminating in conclusions and future research directions.