Key points are not available for this paper at this time.
Large and complex topologies in modern cloud environments really call for factors such as fault tolerance and efficient usage of resources.Current fault detection and load balancing techniques are often found to be insufficient due to known limitations of very high false positives, late detection, and great redundancy overheads that often-become bottlenecks for performance.To this effect, this work offers a new hybrid faulttolerant load-balancing framework with an integration of multiple advanced techniques as follows: Hybrid Autoencoder-Based Anomaly Detection (HAAD), Task-Level Replication Using Intelligent Redundancy Allocation (TRA-IRA) and Long Short-Term Memory (LSTM) networks for proactive failure prediction operations.HAAD discovers known and unknown faults by learning to discern the normal behavior of a system using unsupervised autoencoders, which has achieved 97-98 percent accuracy in fault detection.TRA-IRA dynamically allocates redundant replicas based on task priority and real-time resource health predictions, reducing replication overhead by 20% while maintaining a task completion rate of 99.5%.The LSTM network predicts imminent failures by analysing temporal patterns in system metrics that enable task migration up to 45 min before with 95-96% prediction accuracy.All these techniques are easily integrable with Adaptive ResourceReallocation via Genetic Algorithm (ARR-GA) with respect to optimal scheduling.The Batfly Algorithm is used in an attempt to manage the task.Therefore, due to the integration of these approaches, it presents very efficient performance by increasing by 45% the fault tolerance strength and enhancing the reliability of a system by 50%.The response timestamp along with makespan reduced between 15 to 20%.This model will offer a scalable, dynamic, and robust method of cloud load balancing to augment critical gaps in fault tolerance and optimizations of resources.
Pathania et al. (Mon,) studied this question.