Robust Constrained Optimization for Large-Scale Distributed Computing Resource Allocation

Abstract

Large-scale distributed computing systems call for efficient allocation of heterogeneous resources, involving memory, storage, communication bandwidth and processing capacity. Moreover, conventional resource allocation models generally rely on the assumption that workload requirements and resource availability can be accurately determined in advance. Due to dynamic workloads, node failures, and changing network conditions, variations in parameters are expected in practical cloud, distributed computing, and edge environments. A budgeted robust constrained optimization framework is formulated in the paper. This approach showsuncertain task requirements, where the uncertainty budget parameter controls the budgeted uncertainty set, which maintains stability between efficiency and protection under adverse conditions. The task demand requirements, load-balancing considerations, resource capacity constraints, and allocation costs are efficiently incorporated. Large-scale implementation is addressed through a distributed solution framework guided by the alternating direction method of multipliers. Under convexity assumptions, investigations are made with respect to the robust feasibility, algorithmic convergence, and algorithmic convergence in this study. Mathematical optimisation problems are presented in the paper through the integration of distributed optimization and budgeted robust optimization. When there is variation in the uncertainty levels and system sizes, the deterministic and robust allocation strategies are compared using numerical experiments.