New Algorithm Optimizes Capacity Allocation Across Locations and Service Classes.

A new two-level algorithm has been introduced to efficiently manage a single, conserved capacity budget across numerous locations and two distinct service classes. This system is designed to handle fluctuating and uneven demand, including scenarios where demand surpasses available supply. The first level of the algorithm redistributes capacity within a service class across different locations based on proportional deficits and excesses. The second level then dynamically lends capacity between the two service classes when one has a surplus and the other faces a deficit. The algorithm guarantees exact budget conservation, maintains non-negativity, and achieves a stable allocation in a single iteration under stationary demand, operating with an O(KN) cost per cycle. Its effectiveness was demonstrated in a scenario involving a CDN defending against volumetric attacks, where it prioritized legitimate traffic. The research highlights that a throughput-maximizing objective can be counterproductive under contention, and inter-class borrowing significantly improves high-priority service during bursty loads.