Stability-Preserving Optimization for Adaptive Resource Management in Distributed Computing Networks

Abstract

Dynamic resource allocation in distributed computing networks is challenging because changing workload conditions can cause frequent reconfiguration of task-to-node assignments, even when substantial allocation changes are not necessary. The basic focus of conventional optimization is on the performance of the current allocation and does not explicitly consider the allocation state deployed in the previous decision interval. A trajectory-aware framework, Stability-Preserving Adaptive Resource Optimization using Particle Swarm Optimization (SPARO-PSO), is proposed, which modifies the PSO-based resource optimization framework with temporal allocation stability. The proposed formulation consists of a set of allocation performance measures and an  -norm-based temporal stability measure to discourage generating unnecessary allocations between adjacent allocation states. An adaptive regulation of the strength of stability preservation is provided by a workload-volatility-aware stability coefficient, whereas a bounded allocation transition mechanism restricts excessive reconfiguration. Similarly, the PSO search process is further prolonged via the use of the stability aware fitness evaluation, feasibility repair, and global-best selection with constraints. Theoretical analysis of the framework is performed for startable optimal allocation, bounded transition of allocations, adaptive stability response, stability dominance and stable allocation-region behaviour. Experimental evaluation involves static allocation, Round-Robin, greedy adaptive allocation, conventional PSO and PSO without stability regularization under stable, gradual, sudden-spike and oscillatory workload conditions. Alongside the traditional performance metrics, the following are also employed to measure allocation stability: the Allocation Stability Index, Allocation Oscillation, Reconfiguration Frequency, and Stable Region Convergence Time. The study defines a trajectory-aware perspective on adaptive resource management which takes into account allocation quality, transition behaviour, and temporal stability.