Horizontal Pod Autoscaling in Kubernetes Cluster Using Long Short-Term Memory

Containerization is a major trend in software development as an alternative to virtualization. It involves encapsulation of software code, libraries, packages, and dependencies, so that it can run uniformly and consistently on any infrastructure. On a single server or virtual machine, by using the same operating system, we can deploy multiple application with the help of containerization. One such open-source container orchestration technology that supports declarative setup and automation for managing containerized workloads and services is Kubernetes. It comprises a vast environment that is expanding and evolving quickly. Tools, assistance, and services for Kubernetes are readily accessible. Horizontal pod autoscaling plays a pivotal role in the modern-day world due to diversified user requests and resources. This paper proposes a custom autoscaler to perform horizontal pod autoscaling. The effectiveness of this lstm-based autoscaler is compared with the default autoscaler based on their response times and SMAPE (Symmetric mean absolute percentage error) error value is considered as an evaluation parameter. From the results, it is revealed that the LSTM model performs better in predicting the scaling of pods compared to the ARIMA model.