New edge architecture: semantic serverless and P4 load balancing for intelligent applications
Within the SMARTY project, a new architecture integrates serverless computing (FaaS), P4-programmable networks and semantic communication to improve efficiency and reduce latency in edge environments by jointly considering resource usage and data semantics.
A P4-based Load Balancer performs real-time traffic steering using CPU, GPU and memory utilization, node capabilities and packet-level semantics (e.g., person or vehicle), ensuring requests are routed to compatible and less loaded edge nodes.
The system is built on Kubernetes and Knative, extended with function fusion, inter-component routing and edge affinity to support latency-sensitive workloads such as automotive and cooperative perception scenarios.
For observability, OpenTelemetry collects metrics and traces on resource usage, execution time and end-to-end latency, which are analyzed through Elasticsearch, Elastic APM and Kibana.
The approach has been validated on a three-node Kubernetes/Knative cluster integrating Knative Serving, a P4 load balancer, OpenTelemetry Collector and Elasticsearch, with daemon-set deployment to reduce latency and improve system locality and reliability.
Follow SMARTY for updates:
LinkedIn: https://www.linkedin.com/company/smartyeu/?viewAsMember=true
Website: https://www.smarty-project.eu