Reinforcement Learning-Based Dynamic Resource Allocation in Cloud Computing Environments
1st Bandaru Siva Nagendra 2nd Mamidipalli Samba prasad 3rd Gudipudi Prem Charan
Dept. Computer Application, Aditya University, Surampalem, India bandarusivanagendra239@gmail.com sambaprasad297@gmail.com premcharangudipudi@gmail.com
4th Borusu Jyothi 5th Adapa veera venkata Satyanarayana
Dept. Computer Application, Aditya University, Surampalem, India
jyothiborusu65@gmail.com adapa8827@gmail.com
Abstract—Cloud computing has brought fresh computing landscape by enabling the limitless, extensive and on call access to the computing devices through internet. With the exponential increase in the number of cloud-based applications, which includes and are not limited to real-time analytics systems and artificial intelligence systems, e-commerce websites and health care websites, efficient allocation of resources has become a burning issue . The dynamism and unpredictability of workloads together with the heterogeneity of cloud infrastructures imply that the standard strategies of resource management find it difficult to attempt to provide the best of performance at the minimal cost of operation. More classic approaches, such as systems that use static thresholds and systems based on rules tend to be too inflexible to respond to the rapidly shifting workload trends and often result in over- or under-provisioning of resources
.
A reinforcement learning-oriented dynamic resource allocation can be suggested in this study to solve these issues. The new model offers Deep Q-Network (DQN) to learn the optimal scaling policies that include the experience with the cloud world . Unlike the conventional methods, the reinforcement learning agent dynamically adjusts its decisions that regulate the resource allocation process based upon the real time system states, including CPU utilization, working memory utilization, and workload volume, and response time latency . The rewarding operation value is calculated in a strategic way to take into account a set of various objectives, such as maximization of resource consumption, presence of violations of service level agreement minimization, and minimization of cost of operations. Experimental cloud isosystem modeled analysis draws that the proposed solution has been established to be significantly more efficient, flexible and robust than the longstanding solution. The results indicate the fact that the process of reinforcement learning provides a fresh outlook on intelligent and autonomous cloud resource managing systems that can cater to complex and complex workloads.
Keywords: Multimodal deep-learning, Fake news detection, social signals, transformer models, graph neural networks, multimodal deep-learning.
Index Terms—component, formatting, style, styling, insert
