Dynamic Scaling of Resources for Streaming Applications

Abstract

Cloud computing has transform the way applications and services are deployed and man- aged, offering flexible and scalable resources to meet varying demands. One crucial aspect of cloud computing is resource scaling, which involves dynamically adjusting the alloca- tion of computing resources, such as CPU and memory, to ensure optimal performance and cost efficiency. Streaming applications often require real-time decision making based on the incoming data. Time series forecasting using a transformer-based neural network enables us to predict future values or trends in the streaming data, allowing us to make timely and informed decisions. Streaming applications often require real-time decision making based on the incoming data. Time series forecasting using a transformer-based neural network enables us to predict future values or trends in the streaming data, allowing us to make timely and informed decisions This work provides a time series forecasting implementation of a transformer-based neu- ral network for streaming applications. We explore the concept of dynamic resource scaling in cloud computing and its impact on streaming applications. It predict resource requirements for incoming data. The objective is to investigate how dynamic scaling can enhance the efficiency and cost-effectiveness of streaming applications by allocating resources based on real-time CPU and memory utilization. The implemented method for training a transformer-based neural network using a time series dataset that includes CPU, memory etc. The architecture of the transformer model is design to recognize temporal connections in the data, allowing precise resource estimations. We examine how dynamic scaling affects the functionality of streaming applications. Us- ing evaluation metrics like RMSE, MSE, and MAE, we analyse the prediction model’s accuracy.