IP3 2025 – Memory Circuits (Lot 16328)

This lot is generally related to a hybrid technique that allows dynamic partitioning of memory bandwidth based on application behavior on commodity servers without offline profiling to provide real-time efficient bandwidth use and improve application performance. Disclosed is a technique for partitioning memory bandwidth in a multi-core system by analyzing and profiling application-specific state combinations. The system identifies optimal configurations by defining core allocation and bandwidth settings. Also disclosed are techniques to select a final state by exploring the system’s state space and applying bandwidth partitioning accordingly to improve memory bandwidth utilization, reduce interference among applications, and enhance performance predictability. The technology may be implemented in cloud computing systems, real-time embedded systems, high-performance computing systems, multi-tenant server environments, etc.