Cost-Aware, Profiling-Based Resource Configuration for Virtual Network Function Forwarding Graphs

Network Function Virtualization (NFV) enables Virtual Network Functions (VNFs) to run on shared physical infrastructure, offering flexible resource management. A key challenge in NFV environments is determining the optimal resource configuration for VNFs, particularly in a VNF-Forwarding Graph (VNF-FG) where dependencies between VNFs complicate the process. This paper introduces a Cost-Aware Profiling-based Resource Configuration (CA-PRC) algorithm, which dynamically adjusts the resource configurations to minimize the total allocated CPU while achieving performance targets specified in the Service Level Agreements (SLAs). The proposed CA-PRC algorithm incrementally generates performance profiles for VNFs and balances the trade-off between CPU allocation and profiling cost, thus reducing the need for an exhaustive profiling. Our evaluations demonstrate that the proposed algorithm significantly outperforms the Uniform Random Selection (URS) and Weighted Random VNF Selection (WRVS) in terms of the number of required profiling rounds and total allocated CPU.