eZtunnel: An eBPF-based Traffic Acceleration Mechanism for Cloud-Native Infrastructures

Cloud-native applications, characterized by their scalability, resilience, and flexibility, adopt a microservices architecture to decompose applications into smaller, independently manageable services. This architecture, while offering significant benefits, introduces challenges in service-to-service communication, requiring the use of advanced orchestration and communication frameworks such as Kubernetes and Istio, respectively. However, the complexity of these technologies impose substantial overhead on the underlying infrastructure, introducing longer packet processing paths.

This work identifies the performance bottlenecks arising from the extensive use of service meshes, highlighting the critical issue of CPU overload due to networking-related tasks. To address these challenges, we propose eZtunnel, a transparent offloading technique to allow efficient communications in service meshes. This proposal leverages extended Berkeley Packet Filter (eBPF) as the enabler technology to address and mitigate the problem.

Experiments show that the proposed solution can improve intra-node networking metrics, such as reduction in average Flow Completion Time (FCT) by 41.2%, latency by 42.0%, and increase in throughput by 68.5%. Memory footprint was small, reaching at most 60.5 MB. CPU usage was variable between +29.2% and -23.2%. Through this approach, the research aims to unlock the full potential of cloud-native applications, ensuring that the architectural advancements translate into relevant benefits.