International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 12 | Dec 2024
p-ISSN: 2395-0072
www.irjet.net
Dynamic Resource Allocation in Cloud Computing Using Machine Learning Techniques M V Narayana1, M Vijaya Sudha2, T Sumallika3, S Madhu4, B Mamatha5, Ravikumar Thallapalli6 1Professor, Department of CSE, Guru Nanak Institutions Technical Campus, Hyderabad, Telangana, India
2Assistant Professor, Department of IT, Sir C R Reddy College of Engineering, Eluru, Andhra Pradesh, India. 3Assistant Professor, Department of IT, SR Gudlavalleru Engineering College, Gudlavalleru, Andhra Pradesh, India. 4Professor, Department of CSE(AIML), Guru Nanak Institutions Technical Campus, Hyderabad, Telangana, India. 5Assistant Professor, Department of CSE, Keshav Memorial Engineering College, Hyderabad, Telangana, India. 6Assistant Professor Dept of CSE, Vaageswari College of Engineering, Karimnagar, Telangana, India.
---------------------------------------------------------------------***--------------------------------------------------------------------1.INTRODUCTION Abstract - Cloud computing, being the most promising development in recent years for service delivery platform that enables on-demand access, Scalability and Efficiency to be cost-effective has to deal with efficient resource allocation problem which is essential for optimal use of resources at a low operational cost whilst maintaining quality of service (QoS). Classic resource provisioning methods tend not to adapt dynamically as the workloads and environments in cloud systems are remarkably diverse. As a solution to these problems, we present an adaptive resource allocation framework using off-the-shelf machine learning techniques. It uses predictive analytics to predict demand and allocate resources as required in real-time with minimal latency and high availability. In the demand prediction phase, supervised learning algorithms like Random Forest and Gradient Boosting are utilized, whereas reinforcement learning is employed to dynamically fine-tune resource allocation policies informed by real-time feedback. In simulation experiments with different patterns of workload, extensive experiments were carried out. The experimental outcomes indicate that the proposed architecture achieves a notable enhancement in resource utilization efficiency by decreasing both underprovisioning and over-provisioning cases of up to 30% compared to traditional static approaches. Furthermore, the framework demonstrates improved scalability against workload variations and guarantees that QoS metrics like response time and throughput remain stable. In addition to improving both scalability and reliability of cloud infrastructures, sustainability is achieved by reducing energy consumption. We also conducted a detailed analysis for the per-instance computation cost of incorporating machine learning models into certain typical cloud orchestration processes, demonstrating that this approach is practical on real cloud usage. Through its complexity and adaptability, this study contributes towards the state-of-the-art of cloud resource management. These results will be valuable for highlevel insight to develop efficient, adaptable, and sustainable cloud computing systems for industry players and academics.
Cloud computing has transformed access to accessible and functional computing by providing zoned space to meet business needs changing over time. This dynamism has created an unprecedented need for automated resources management approaches that can be tuned dynamically to a wide variety of workload traces. We focus on cloud computing since in cloud computing the cost of resources is comparatively higher and resource allocation plays a key role by distributing computational resources, such as CPU, memory, storage and network bandwidth to different applications and users to use resources efficiently. Unfortunately, traditional resource allocation mechanisms may not be sufficient for the increasingly complex and dynamic nature of modern cloud environments with high variability, heterogeneity and scalability requirements. Dynamic resources allocation intends to maximize the performance of cloud infrastructure by constantly observing and responding in real-time to changes in workload, user demands, and hardware limitations. Effective allocation strategies are essential to preserve quality of service (QoS), cut operation costs and energy consumption. Yet, meeting these goals is non-trivial due to workloads' unpredictable nature, the several types of cloud applications (box-job versus boxes-job) and the dynamic interactions between resource constraints. At the same time, as opposed to supervised learning which requires labeled data, reinforcement learning allows systems to learn via real-time feedback; a feature that makes it very advantageous for dynamic and adaptive resource management. Reinforcement learning excels at discovering appropriate policies by iteratively navigating through and exploiting various allocation strategies, thus identifying optimal resource utilizations paired with QoS metrics. Benefits of Integrating ML Techniques into Resource Allocation Frameworks FRST, ML models can learn from evolving workload patterns, infrastructure configurations, and user requests to provide performance benefits in all conditions. ML-based approaches can also lower operational costs by reducing the cost of over-provisioning and under-
Key Words: Dynamic Resource Allocation, Cloud Computing, Machine Learning, Predictive Analytics, Quality of Service.
© 2024, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 221