i-Profiler: Towards Multi-Objective Autonomous VNF Profiling with Reinforcement Learning

The intelligent orchestration of Virtual Network Functions (VNFs) requires understanding and profiling the impact of VNF resource consumption on its performance. Recent efforts focus on ML model-based profiling methods to discover and manage optimal trade-offs between cost-efficient resource combinations for VNFs and the latter's performance. To this end, the current paper poses a novel effort towards an intelligent VNF profiling by targeting multiple resource and performance optimisation objectives, thus suiting real-world applications. Our approach is based on adapted Reinforcement Learning (RL) considering three types of resources: CPU, memory, and network link capacity, as well as the output load and performance of VNFs. Our current results show how we can improve the VNF performance while at the same time optimising the consumption of multiple resources in contrast to single-objective solutions in the literature. We investigate a VNF type via exhaustive resource and performance profiling against our intelligent adapted RL approach. In addition, as a benchmark model to RL, we compare our model with a Supervised Learning (SL) model. Our results denote successful profiling decisions with greater resource prediction accuracy, paving the way for future research.