A multi objective DB-RNN based core prediction and resource allocation scheme for multicore processors
P.R. Mohanan, Mariamma Chacko · Computers & Electrical Engineering · 2024
Asymmetric Multicore Processors (AMP) are widely used in both advanced and basic computing systems due to their fundamental flexibility and outstanding computing possibilities. Among the subcategories of AMP, Performance Asymmetric Multicore (PAM) Architectures are unique in that they include various micro-architecture cores into a single chip . Nevertheless, the complex interplay between heterogeneous cores and a diverse range of applications presents a formidable challenge in determining the optimal hardware configuration , encompassing core selection and reduced energy consumption for each application. To tackle these multifaceted problems, this paper introduces a pioneering model for core prediction and resource allocation based on the Dual Branch Recurrent Neural Network (DB-RNN), specifically for AMP. The DB-RNN model encompasses weight sharing facilitated by a hybrid optimization algorithm named African Vulture with Aquila Optimizer (AVAO), and a core prediction module. The proposed model evaluates the Energy-Delay Product (EDP) to find the performance of each core. In the final phase, DB-RNN dynamically predicts the most suitable cores for individual workloads at runtime, thereby elevating both energy efficiency and overall system performance . The experimental results show that the model achieved prediction accuracy up to 98.76 %. This innovative approach paves the way for enhanced efficiency and performance in PAM systems .