A Framework for Compute-Aware Task Placement and Traffic Steering in Heterogeneous Computing Networks

The increasing deployment of geographically distributed computing infrastructures equipped with heterogeneous compute resources (e.g., CPU, GPU, memory) has motivated new architectural approaches for jointly optimizing task placement and traffic steering. In heterogeneous computing networks, tasks must be assigned to compute- capable nodes while respecting multi-dimensional resource constraints and network bandwidth limitations. This document presents a conceptual framework for Compute-Aware Task Placement and Traffic Steering (CATPTS). The framework models a computing network as a directed graph containing compute-capable nodes and forwarding-only nodes. Task execution location selection and two-stage traffic steering are jointly optimized under link bandwidth and multi-dimensional compute capacity constraints. The objective is to achieve global load balancing across compute and network resources. This document defines the architectural principles, conceptual model, terminology, and optimization formulation underlying such systems. It does not specify protocol mechanisms.