Kubernetes container orchestration technology is beginning to address telco Network Function Virtualization (NFV) deployment needs of: i) multiple data-plane interfaces and one management-plane interface, and ii) support for various protocols/software stacks targeting I/O performance, including software acceleration technologies and software switch architectures. The purpose of this work is to accurately benchmark different network-virtualization architectures (OvS-DPDK, SR-IOV and VPP) in a Kubernetes deployment for telco data plane configurations. By leveraging industry-standard testing methodologies and an open source testing framework (VSPERF) that can control different hardware-based commercial and open source traffic generators, we benchmark the data plane of the telco NFV deployments. We also study the effects of different resource allocation strategies (CPU-Pinning, compute and storage resource variations, NUMA) on the data plane performance.
Benchmarking Kubernetes Container-Networking for Telco Usecases
Federica Paganelli;
2021-01-01
Abstract
Kubernetes container orchestration technology is beginning to address telco Network Function Virtualization (NFV) deployment needs of: i) multiple data-plane interfaces and one management-plane interface, and ii) support for various protocols/software stacks targeting I/O performance, including software acceleration technologies and software switch architectures. The purpose of this work is to accurately benchmark different network-virtualization architectures (OvS-DPDK, SR-IOV and VPP) in a Kubernetes deployment for telco data plane configurations. By leveraging industry-standard testing methodologies and an open source testing framework (VSPERF) that can control different hardware-based commercial and open source traffic generators, we benchmark the data plane of the telco NFV deployments. We also study the effects of different resource allocation strategies (CPU-Pinning, compute and storage resource variations, NUMA) on the data plane performance.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
