An Enhanced Hybrid Task Scheduling with Load Balancing in Cloud with Prediction of Future Resource Demands

Cloud Computing (CC) faces major challenges in Task Scheduling (TS) and Load Balancing (LB), especially in dynamic resource allocation for different tasks in heterogeneous environments.CC systems have implemented many metaheuristic algorithms for TS-LB over the past decades.Among many others, a hybrid TS-LB using the Dove Optimization Algorithm (DOA), called TSLBDOA, achieved minimum makespan and maximum load fairness.This algorithm can simultaneously pick optimal tasks and Virtual Machines (VMs) to be migrated to the appropriate Physical Machines (PMs) based on multiple objectives, such as the usage of CPU, memory, bandwidth, and storage of all VMs and PMs.However, fluctuations in resource allocation and heterogeneity of cloud resources can negatively influence TS-LB, leading to high resource wastage and poor Quality of Service (QoS) requirements.To tackle this problem, cloud applications should dynamically allocate resources based on the user's requirements, ensuring that the number of allocated resources equals the quantity required.Hence, this article introduces an Enhanced Hybrid TS-LB with Prediction of Future Resource Demands (EHTSLB-PFRD) for CC.This algorithm incorporates the Gated Recurrent Unit (GRU), DOA, Analytic Hierarchy Process (AHP), and Petri-Recurrent Wavelet Fuzzy Neural Network (PRWFNN)-based predictive model for dynamic resource allocation in heterogeneous CC and to satisfy end users' requirements effectively.The GRU is used to predict task execution time, which assists in categorizing tasks as dependable and undependable.Besides, the PRWFNN is trained to forecast upcoming resource demands at various intervals based on the user requests and historical resource utilization data.Thus, this model predicts the resource requirement, allowing for the timely allocation of sufficient resources to perform the user's request.Moreover, the predicted outcomes can assist the DOA to optimize the scheduling of dependable tasks, whereas the AHP is used to backfill the undependable tasks.This DOA is also used to choose the optimal VMs for migration and achieve LB.Finally, simulation results validate that this EHTSLB-PFRD achieves a 2430 sec makespan, a 0.98 Mean Resource Utilization Ratio (MRUR), a 0.04 Degree of Imbalance (DoI), a 0.93 tasks/sec throughput, 21 MB memory usage, 167 MB/sec bandwidth usage, 59 kWh energy usage, and 1.28 load fairness for 1000 tasks, compared to the standard TSLB algorithms in CC systems, such as First Come First Serve (FCFS), Round Robin (RR), Shortest Job First (SJF), and Max-Min.