NURPS: NUMA- and Reliability-aware Parallelized SFC Deployment in Multi-Core Servers

Network Function Virtualization (NFV) leverages Virtualized Network Functions (VNFs) to overcome the limitations of traditional middleboxes by using standard hardware to create Service Function Chains (SFCs) that deliver specific services. However, virtualization overhead and SFC length can contribute to increased end-to-end delay. To address this, Paral-lelized SFC (PSFC) enables the simultaneous execution of identical packets by multiple VNFs, significantly reducing delay. Most existing SFC placement approaches assume uniform treatment of all CPU cores in multi-core NFV servers, overlooking the impact of Non-Uniform Memory Access (NUMA) architecture. Studies have shown that arbitrary VNF placement can degrade SFC performance due to resource contention within a NUMA node—specifically in the Last Level Cache (LLC) and Memory Bandwidth (MB)—and between NUMA nodes, particularly in Quick Path Interconnect (QPI) bandwidth. This paper presents NURPS, a NUMA-aware PSFC placement strategy that deploys VNFs by allocating critical resources, such as LLC and MB, across NUMA nodes to maximize performance while minimizing server usage. Furthermore, NURPS incorporates both software and hardware reliability into the VNF deployment process, ensuring stable performance under varying operational conditions. Extensive simulations demonstrate that NURPS effectively reduces the number of servers required for PSFC deployment while maintaining high reliability.