AUTONOMOUS CLOUD RESOURCE OPTIMIZATION USING REINFORCEMENT LEARNING FOR FINTECH MICROSERVICES

Abstract

The fintech sector's quick adoption of microservices architectures has led to resource cloud management issues never seen before. The older autoscaling models that manage the machines in areas of computing demand are based on the policy of assuming certain threshold values and therefore being reactive in their response. Such traditional policies fail to predict demand accurately and result in either server resources being wasted or services being interrupted. Thus, this research suggests an intelligent cloud resource management system based on reinforcement learning which will tune the compute power for microservices in the FinTech industry. The system consists of three main parts: an RL-based autoscaler that discovers the best scaling policies through a constant interaction with the production workloads, an NLP-based infrastructure code validator that inspects the Terraform and Ansible scripts and gives the go-ahead or not for the deployment, and a deployment risk predictor that uses the past data and anomaly classification. Analyzing the FinTech workloads systematically, we are able to show that RL-based autoscaling not only cuts infrastructure costs by 31% but also increases response time variability by 42% when compared with conventional methods. Moreover, the NLP validator is capable of detecting configuration mistakes with 89% accuracy and thereby eliminating the chances of deployment failures. Additionally, the model that predicts risks can flag deployments with 85% accuracy in terms of those likely to pose high risk before their execution. All these results point to a significant impact on organizations in the financial sector that are trying to maintain operational efficiency, reliability, and compliance in the cloud without incurring extra costs.

DOI: 10.46121/pspc.53.3.15