Answers are grounded in this paper and cite their sources.
Mobile edge–cloud computing (MECC) must reconcile periodic capacity preparation with short-timescale workflow decisions under variable demand and finite resources. We propose a two-timescale proactive–reactive framework in which deterministic forecasting and fixed-stream Monte Carlo evaluation select a finite-set, expected-profit-maximizing capacity threshold at the macro timescale. The cached capacity plan then conditions categorical multi-agent workflow offloading at the micro timescale; reactive cloud fallback corrects residual infeasibility, and sub-deadline urgent priority (SUP) scheduling operates through a separate reconstructed-DAG path. The framework is solver decoupled: adapted multi-agent deep deterministic policy gradient (MADDPG) is the default offloading solver, while adapted multi-agent proximal policy optimization (MAPPO) is integrated through the same interface. For the provisioning module, we prove consistency of the finite-candidate Monte Carlo estimator, finiteness of its integer candidate/output domain, and compactness of the continuous relaxation. A layer-matched validation design connects these analytical properties to controlled component comparisons, frozen-policy held-out tests, sensitivity studies, and an independent scheduling benchmark. Across five independently trained seeds, the complete MADDPG configuration achieves a held-out mean latency of 8.675 ms, a deadline-miss ratio of 26.52%, and a team return of 307.12. Nominal equal-budget results favor prediction-augmented fixed-capacity MADDPG in latency and tardiness, whereas under equal-mean capacity stress Forecast–MC yields favorable point estimates for four of five metrics; the paired confidence intervals nevertheless cross zero. Component removal confirms that learned offloading contributes beyond the installed capacity prior and that reactive fallback preserves feasible cooperative execution. The Alibaba-derived workload construction, processed inputs, code, configurations, and reproduction instructions are available at https://github.com/silkyrose/Proactive-MECC-Orchestration .
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: