Using readily-available and fully supported software and configuration options, service providers can now deploy production-ready NFV deployments with deterministic performance and isolation from other VNFs running in the system. We showcase in real-time the consequences of the NFV performance tuning and the trade-off between elastic resource scheduling versus deterministic reserved capacity, and the effects of increased CPU load and network throughput and latency.
A demonstration will be shown comparing the performance of vEPC components (eNodeB and S-GW) running simultaneously on two identical hardware platforms with dual-socket CPUs and SR-IOV cards: one with the generic all-purpose OpenStack configuration, and other a NFV-optimized OpenStack with CPU-pinning, Huge-pages and NUMA affinity.