NUMA-Aware Parallelized Service Function Chain Deployment in Multi-Core Servers

Network Functions Virtualization (NFV) deploys Vir-tualized Network Functions (VNFs) to overcome traditional mid-dlebox limitations. These VNFs, installed on standard hardware, form a Service Function Chain (SFC) to deliver specific services. However, the virtualization layer and the SFC length increase end-to-end delay. To address this, parallelized SFC (PSFC) runs multiple VNFs parallelly, significantly reducing the delay. Previous studies on SFC placement assumed equal treatment of all CPU cores in multi-core NFV servers, overlooking non-uniform memory access (NUMA) architecture. Recent findings demonstrate that randomly placing VNFs can degrade SFC performance due to resource contention at intra-node and inter-node levels. In this paper, we propose a NUMA-aware approach that strategically allocates resources like Last Level Cache (LLC) and Memory Bandwidth (MB) to minimize required servers and reduce PSFC overheads. Extensive simulations demonstrate that the proposed approach outperforms in terms of servers needed for PSFC deployment while ensuring performance.