Integration of a Container Network Interface (CNI) plugin for 6G high performance shared memory communication in Kubernetes

Abstract

As distributed applications increasingly rely on containerized deployments and service oriented communication models, conventional TCP based communication can introduce additional overhead for co-located workloads operating within the same pod or node. The study examines how shared memory resources can be integrated into Kubernetes environments through preload based interposition while preserving compatibility with existing TCP oriented applications and workflows. This thesis presents and evaluates two integration architectures, namely the sidecar architecture and the embedded architecture. The sidecar approach externalizes coordination into a dedicated container within the pod, while the embedded approach integrates coordination directly into the preload library through an ownership model. Both architectures retain the conventional socket based programming model from the perspective of the application, enabling acceleration without requiring modifications to application source code. The proposed implementations are evaluated experimentally across multiple Kubernetes environments, including KindNetD, Calico, and RKE2. The evaluation focuses on throughput behavior, connection establishment characteristics and capacity behavior under different payload configurations. The results show that the TCP baseline was mainly concentrated between 220 and 290 Mbps, while SHML enabled deployments increase the throughput, reaching approximately 2100 to 2250 Mbps in the sidecar KindNetD setup, 1250 to 1310 Mbps in the sidecar Calico setup, 1150 to 1250 Mbps in the embedded Calico setup, and around 1160 Mbps in the embedded RKE2 setup. Connection establishment remained within a narrow millisecond range across the evaluated deployments, while capacity testing showed that 32 KiB and 64 KiB payload configurations produced the most concentrated high throughput results, with most throughput observations grouped between approximately 70 and 85 Gbps.