Sustainable Data Dependency Resolution Architectural Framework to Achieve Energy Efficiency Using Speculative Parallelization

Parallelization algorithms have made remarkable progress since the introduction of speculative parallelism for automatically parallelizing tasks on multi-core chips. This research focuses on the difficulties associated with developing adaptable architecture and enhancing energy efficiency in parallel computing systems, with a strong emphasis on sustainability. A sustainable data dependency resolution framework that leverages speculative parallelization and the LLVM compiler infrastructure for portability is proposed. This approach focuses on analyzing and minimizing inter-dependencies to improve performance and reduce energy costs, thus promoting sustainable computing practices. Additionally, the optimal scalar voltage scaling algorithm and optimal dynamic voltage scaling mechanism, aided by profiling tools, minimize energy consumption while preserving speculative parallelism. Experimental results demonstrate a significant reduction in energy consumption, achieving a sustainability milestone of 71.03% compared to pure speculative parallelism, with a modest increase in the geometric mean speedup of 1.71x.