ETD

• cats
• centralized
• orchestration
• steering
• traffic

The goal of this thesis is to design, implement, and evaluate a dynamic Computing-Aware Traffic Steering (CATS) framework to ensure intelligent load balancing across a distributed network of computing resources in the edge-cloud continuum. The proposed solution creates an orchestration system that supports continuous adaptation to variable traffic loads and changing server-side computing capabilities, leveraging carrier-grade networking technologies.
The system is architected around a centralized orchestrator, linked to a BGP Route Reflector, which maintains a real-time view of the entire network ecosystem.
To achieve real-time network state awareness, the orchestrator utilizes the BGP Monitoring Protocol (BMP).
On the Ingress routers, incoming traffic is classified based on service-level requirements.
The core of the decision-making process is a centralized min-max fairness algorithm, implemented in C++, which is executed by the orchestrator. This algorithm takes as input the predicted traffic demands from each Ingress and the available computing capacities from each service instance at the Egress.
Upon calculating the optimal traffic distribution the orchestrator translates these decisions into network policy. It achieves this by communicating with agents deployed on the Ingress routers, to install in their tables a value for each route, used in the configured Weighted Equal-Cost Multi-Path (WECMP). This enables them to distribute traffic across multiple available paths.