capacity allocation design

This paper addresses the challenge of allocating limited network resources under persistent, adversarial attacks with unknown statistical properties. Method: We propose a bi-objective optimization framework that jointly minimizes system-wide damage and long-term resource scheduling and inter-node relocation costs. Our approach innovatively integrates chance-constrained programming with dynamic network flow optimization; it decomposes the problem to coordinate node-level defense configuration and global resource allocation, and incorporates online learning for adaptive attack pattern identification. Contribution/Results: We theoretically prove that the method converges to a near-optimal solution without prior knowledge of attack distributions. Experiments across diverse attack intensities and network topologies demonstrate that our method significantly outperforms three baseline strategies—reducing average system damage by 32.7% and operational cost by 28.4%. The framework exhibits strong robustness and practical applicability in dynamic, uncertain adversarial environments.

This study addresses the challenge of optimizing resource allocation for multi-hazard risk mitigation in U.S. homeland security and emergency management. We propose an integer linear programming (ILP) decision-support model that integrates probabilistic risk assessment (PRA) with multi-criteria consequence quantification. Methodologically, the model innovatively fuses heterogeneous historical data and publicly available information to enable joint modeling across 16 hazard types and six consequence dimensions, while selecting optimal mitigation projects under budget constraints. It further introduces a sensitivity-driven framework that jointly optimizes robustness and cost-effectiveness. Applied empirically in Iowa, the model generates a high-value portfolio of 52 mitigation projects, achieving an average 37% reduction in expected risk. Multi-scenario sensitivity analysis confirms solution robustness. The approach provides a scalable, methodologically rigorous foundation for evidence-based resilience investment.

This study addresses the limitation of traditional cascade failure models, which assume binary node states and thus fail to capture the partial functionality commonly observed in real-world multilayer systems. To overcome this, the authors propose a multilayer flow network model incorporating partial node functionality and introduce an interlayer influence factor to characterize cross-layer resource competition. Leveraging mean-field theory, they derive recursive equations that predict the fraction of surviving nodes in each layer after cascade termination. This work presents the first systematic modeling of cascade dynamics under partial functionality, uncovering novel phenomena such as non-monotonic robustness curves and equivalent single-layer survival phases. Furthermore, the authors develop a capacity allocation strategy that integrates interlayer influence with local topological information, demonstrating significantly improved system robustness over baseline methods under a fixed total capacity budget.

This work addresses the challenge of maintaining network-wide connectivity in communication networks where link lengths vary over time and node connectivity is subject to uncertainty. To ensure robustness under worst-case scenarios while minimizing regenerator deployment costs, the paper proposes a novel robust optimization approach. Its key innovation lies in the construction of a dynamic budget uncertainty set that effectively captures temporal and structural uncertainties, combined with an integrated solution framework leveraging column-and-constraint generation, Benders decomposition, and a learning-enhanced hide-and-seek game to improve both model adaptability and computational efficiency. Theoretical analysis and extensive experiments demonstrate that the proposed method significantly outperforms conventional static robust models and deterministic worst-case approaches, achieving guaranteed connectivity with substantially reduced deployment costs.

This study addresses the regenerator deployment problem in fault-tolerant optical networks under link failures and signal attenuation constraints, explicitly incorporating discrete uncertainty in regenerator installation costs—a consideration previously unexplored in the literature. The authors propose a robust optimization approach that guarantees full network connectivity under the worst-case cost scenario. Building upon a flow- and cut-based integer programming framework, they develop two exact solution algorithms that integrate robust optimization theory to handle cost uncertainty. Theoretical analysis and computational experiments demonstrate that the proposed method efficiently computes optimal deployment strategies across diverse uncertainty scenarios, effectively balancing cost control with guaranteed network connectivity.

Traditional scaling law estimation suffers from high computational costs due to the absence of efficient budget allocation strategies. This work proposes a novel approach that, for the first time, integrates surrogate-guided pruning into scaling law modeling by combining the Successive Halving algorithm with both parametric and non-parametric surrogate models. This integration enables proactive allocation of computational resources and efficient construction of loss-compute Pareto frontiers. The method substantially improves resource utilization efficiency, achieving relative performance gains of up to 2.84% on real datasets and 5.47% on synthetic datasets, while reducing computational costs by as much as 98.7%.

This work addresses dynamic, heterogeneous, and potentially budget-exceeding demand in multi-location dual-service scenarios by proposing a two-level adaptive capacity allocation algorithm. The first level proportionally redistributes surplus and deficit capacities within each service class across locations, while the second level enables elastic cross-class capacity borrowing to handle bursty loads. This approach uniquely integrates intra-class proportional reallocation with inter-class elastic borrowing, achieving stateless, single-round convergence under a fixed budget—thereby overcoming throughput maximization limitations in contention-prone settings. With computational complexity O(KN) for K service classes and N locations, the algorithm supports real-time scheduling. Experiments demonstrate that, in CDN-based defense against traffic attacks, it satisfies 66%–93% of high-priority requests—matching the performance of single-class linear programming optima—while consistently avoiding both underutilization and over-provisioning even when total demand exceeds the budget. A prototype validates its efficacy under real HTTP traffic.

This work addresses the low bandwidth utilization and high transmission latency in high-performance computing centers caused by static allocation and simplistic queuing. To overcome these limitations, the authors propose a scientific-value-driven dynamic bandwidth allocation mechanism that jointly models network and computational constraints. Users participate in resource allocation through XOR bids specifying their data requirements and associated scientific value. The approach introduces two novel auction mechanisms—the greedy value-density auction and the VCG knapsack auction—balancing practical efficiency with theoretical optimality. Experimental results demonstrate that under high load, the proposed method reduces both average and tail task completion latency by over 80% compared to first-come-first-served scheduling, decreases the coefficient of variation in latency by 75–85%, and lowers the peak-to-mean network load ratio by 60–70%, substantially enhancing system stability and resource utilization efficiency.

This work addresses the problem of optimally allocating capacity to network nodes under a fixed budget to mitigate cascading failures triggered by local load redistribution. The authors propose TANGCO, a novel approach that, for the first time, integrates graph neural networks with policy gradient reinforcement learning, enabling end-to-end training within a cascading failure simulator. A heuristic anchoring mechanism is introduced to facilitate topology-aware capacity allocation. TANGCO supports cross-graph transferability, allowing pretrained models to be deployed on new networks without fine-tuning. Evaluated across 450 synthetic and 45 real-world networks, the method significantly outperforms existing heuristics on 40 of the real networks, improving robustness by 1.6% to 246%. The training scales nearly linearly with network size, and deployment incurs computational costs comparable to those of heuristic baselines.