Hybrid Fuzzy Spark Lion Whale Algorithm for Energy-Efficient Resource Allocation and Task Migration in Cloud Data Centers
TL;DR
The results empirically quantify the necessary tradeoff between aggressive hardware consolidation and Service Level Agreement preservation, establishing Fuzzy-SLW as a scalable solution for power-constrained hyper-scale environments.
Abstract
: The exponential growth of cloud data centers necessitates highly efficient resource allocation and task migration strategies. However, multi-dimensional memory fragmentation severely limits the efficacy of standard scheduling algorithms under heavy-tailed, real-world workloads. This paper proposes Fuzzy-SLW, a hybrid swarm-intelligence architecture that integrates a Mamdani fuzzy-inference pre-filter with a distributed Spark Lion-Whale Optimization (SLWO) core via Apache Spark. The fuzzy pre-filter mathematically prunes the search space using non-compressible hardware constraints, while the Spark execution model resolves the traditional serial bottleneck of swarm intelligence. Evaluated within a discrete-event environment utilizing the Google Cluster Trace (2019), Fuzzy-SLW demonstrates a greater than 240% relative improvement ( + 42.2 percentage points) in virtual machine utilization over load-scattering metaheuristics and avoids the premature policy convergence observed in Deep-DQN baselines. For large-population offline optimization configurations ( P ≥ 5000 individuals), the distributed architecture achieves a 5.85 times sub-linear Amdahl speedup; below this population threshold, including the P = 20 configuration used for online, per-task scheduling, thread-pool context-switching overhead dominates and distributed partitioning does not improve wall-clock latency. The results empirically quantify the necessary tradeoff between aggressive hardware consolidation and Service Level Agreement preservation, establishing Fuzzy-SLW as a scalable solution for power-constrained hyper-scale environments.