A Hybrid Metaheuristic Model for Multi-Objective Optimization in Distributed Computing Networks

Abstract

This paper proposes a hybrid metaheuristic framework to jointly optimize latency, throughput, energy consumption, and reliability in distributed computing networks spanning edgefogcloud tiers. The model integrates a decomposition-based global search with adaptive multi-objective evolutionary operators and a problem-aware local intensification stage. Specifically, a population is evolved using a Pareto-dominance engine with adaptive operator selection (crossover/mutation portfolios inspired by NSGA-II and differential evolution), while an ant-colony/gradient-free local search refines elite solutions near congestion-prone regions of the network. To reduce evaluation cost under dynamic workloads, we incorporate surrogate fitness modeling via incremental regression on sampled flows and a feasibility repair mechanism that respects resource, SLA, and security constraints. A restart-and-memory strategy preserves diversity when fronts stagnate and accelerates convergence after topology or demand shifts. Extensive simulations on heterogeneous topologies with time-varying traffic demonstrate superior convergence speed, improved hypervolume, and better spread of Pareto fronts compared to canonical MOEAs and single-method hybrids. The approach consistently yields lower end-to-end latency and energy budgets at comparable reliability particularly in mixed CPU/GPU clusters and bandwidth-constrained edge segments. Ablation studies confirm the contribution of (i) adaptive operator selection, (ii) surrogate-guided evaluations, and (iii) local intensification. The proposed framework offers a practical pathway for online, policy-driven orchestration in large-scale distributed systems, enabling operators to trade off QoS and cost under uncertainty