Date of Publication :30th November 2023

Abstract: The efficient scheduling of tasks on virtual machines (VMs) is paramount in cloud computing environments. The complexity and dynamism of today's applications require a more insightful and adaptive approach to task allocation to ensure optimal resource utilization and service delivery. Traditional scheduling approaches often fall short when it comes to considering the multi-dimensional attributes of tasks and VMs, such as makespan, deadline, memory, and bandwidth requirements. These methodologies lack the ability to dynamically adapt to the ever-evolving requirements of tasks and the capacities of VMs, leading to suboptimal performance and resource wastage. In this paper, we present a novel approach that fuses BiLSTM & BiGRU with Exponential Smoothing Recurrent Neural Network (ES-RNN) to create a more robust and adaptive task scheduling mechanism under real-time scenarios. This model holistically assesses task capacity based on its makespan, deadline, memory, and bandwidth requirements. Similarly, VM capacity is evaluated based on its RAM, MIPS, bandwidth, and the number of processing elements. The fusion of these advanced neural architectures provides a deeper understanding of the task-VM mapping, enabling a more intelligent and efficient scheduling decision. Our approach demonstrates a marked improvement over traditional techniques, with tangible benefits such as reduced makespan by 4.9% and improved VM computation efficiency by 3.5%. The practical implications of our methodology are profound. By integrating our model into real-world cloud environments, organizations can expect to see an enhanced deadline hit ratio by 1.5%, ensuring that critical tasks meet their time-sensitive objectives. Moreover, the decision-making process becomes significantly more agile, resulting in a decision delay reduction of 4.5%, thereby promoting more responsive and efficient cloud computing operations. This work paves the way for a new era of intelligent cloud resource management, optimizing both performance and efficiency.

Reference :