Abstract

Modern service provider networks must support growing 5G and emerging 6G traffic while making efficient use of finite compute and memory resources. As infrastructure requirements expand, improving workload density and resource utilization can help operators increase capacity while managing capital and operational costs.

F5, Inc. commissioned Tolly to evaluate the infrastructure efficiency and performance of F5 BIG-IP Cloud-Native Edition (CNE) delivering DNS and L4+CGNAT services as cloud-native network functions. Tolly compared CNE with a tested virtualized network function (VNF) deployment in a Red Hat OpenShift environment running on bare-metal Dell PowerEdge systems. 

Tolly measured DNS query throughput and latency using dnsperf and evaluated L4+CGNAT performance using iPerf3. In addition to application performance, Tolly measured CPU and memory consumption to quantify the infrastructure efficiency of each deployment model. Both CNE and VNF configurations were allocated four CPUs and 8GB of RAM, providing a consistent basis for comparison. 

Testing demonstrated significant resource-efficiency advantages for BIG-IP Cloud-Native Edition. For DNS processing, CNE sustained 64,985 queries per second, nearly 1.9× the VNF throughput, while consuming 91.5% less CPU. CNE delivered 123× more DNS queries per GB of memory and reduced average DNS latency by 48%. For L4+CGNAT, CNE delivered comparable throughput, 3,064 Mbps versus 3,177 Mbps for the VNF, while consuming 79.7% less CPU and 98.5% less memory. CNE also reduced p95 L4+CGNAT latency by 82%. 

The results demonstrate that F5 BIG-IP Cloud-Native Edition can deliver DNS and L4+CGNAT services with substantially lower CPU and memory requirements than the tested VNF deployment while maintaining strong application performance, enabling service providers to make more efficient use of Kubernetes infrastructure.