Patent

This patent covers a novel resource allocation method designed to handle variable workloads in AI-driven computing systems. The invention addresses a core challenge in modern AI deployments: efficiently distributing computing resources when workload demands fluctuate unpredictably across real-time tasks.

The method introduces an adaptive mechanism that continuously monitors workload patterns and reallocates processor and memory resources accordingly — reducing idle resource waste while maintaining real-time scheduling deadlines. The approach is particularly applicable to IoT systems and edge AI environments where resource budgets are tightly constrained.

This work builds directly on a body of research into processor dynamic voltage scaling (DVS) and non-volatile memory allocation strategies developed at Sun Research Lab, resulting in a unified technique that is both implementable on commodity hardware and demonstrably effective across variable AI task profiles.