A Low-Discrepancy and Latency-Aware, Scenario-Sensitive Resource Allocation Approach for Cloud Systems Using Latin Square-Based Improvised Genetic Optimization (LSBGO)

Efficient resource allocation in cloud networks is a complex challenge, demanding adaptive strategies to sustain processing millions of tasks whose CPU, memory, and priority profiles changes rapidly. Static heuristics, such as Round Robin and Weighted Least Connection, react quickly but neglect the heterogeneity of resources, thereby creating hotspots and long-tail delays. Meta-heuristics, although excellent, tend to converge slowly and are thus not well-suited for bursty real-time traffic. Hence, the core problem is how to place tasks while ensuring that no virtual machine begins its execution overloaded with work, and how to adjust this placement on the fly as demand, failures, and scale changes occur. This paper presents Hybrid Latin Square-Based Genetic Optimization (LSBGO), a novel framework that integrates deterministic Latin square-based task placement with scenario-sensitive genetic optimization to enable robust, low-discrepancy scheduling across scalable cloud environments. The deterministic Latin-square initializer generates a mathematically uniform random initial distribution over the spatial and temporal task grid. An adaptive genetic phase then evolves the mapping based on live feedback about load and priority, supporting quick rebalance, elastic growth, and failure recovery with minimum tuning of parameters. The framework is extensively evaluated using both steady-state scenarios driven by Google Cluster-Usage traces and stress tests using real-world PlanetLab-style CPU bursts replayed in CloudSim. LSBGO keeps SLA violations below 4 %, maintains a mean time to recovery of 6.9 s, and sustains more than 90 % CPU utilization while holding latency up to 45 ms. The results confirm that LSBGO delivers scalable, low-complexity scheduling that harmonises utilisation, fairness, and responsiveness, making it well-suited for next-generation cloud resource management.