Constraint-Aware Resource Allocation: A Computational Framework for Practical Optimization

Abstract

Computational systems, where there is limited capacity, competing workloads and multioperational constraints, require efficient allocation of resources. The study suggests a constraintaware computational model to evaluate the resources state and aid feasible resourceallocation choices. The framework puts into consideration the CPU, memory, bandwidth, workload nature, task priority, service requirements, deadlines, budget, energy consumption and fairness. Resource-pressure indicators are constructed to measure the demand in comparison with available capacity. The Python analytical workflow that includes preprocessing, feature engineering, exploratory, and supervised machine learning is executed. Accuracy, Precision, Recall, F1-score, ROC AUC and confusion matrices are used to evaluate the performance of the Logistic Regression, Random Forest and Gradient Boosting classifiers. The best overall performance was achieved by the Logistic Regression with 96.6% accuracy, a weighted F1- score of 0.9663, and a ROC-AUC of 0.9994. The feature-importance analysis also showed that CPU, memory and bandwidth pressure are significant predictors of constraint severity. The results show that resource-pressure evaluation in conjunction with predictive classification can be viable in differentiating the various constraints states and also offer a computational foundation to allocate, monitor, and resource-management solutions of constraints in scalable computing settings.