Workload Forecasting Model Using Adaptive Sliding Window in the Cloud Environment

Efficient management of resources in a cloud computing environment heavily relies on workload forecasting. Accurately predicting the workload is critical to ensure optimal utilization of resources. User demands are dynamic and unpredictable, necessitating accurate workload forecasting to allocate resources optimally, ensure service-level agreements, and control costs. Current prediction techniques rely on fixed-size observation windows, which cannot identify local patterns in the most recent data and hence cannot yield reliable findings. Thus, the methods trained with relatively few observations on large, fixed sliding windows lead to wrong estimates or accepting wrong estimates because small windows on changing trends quickly worsen calculations. This research study proposes a proactive workload forecasting model using a Bi-LSTM algorithm with an adaptive sliding window to handle the dynamic and unpredictable nature of user requests (workloads). A proactive workload forecasting model needs to predict incoming future workloads for effective resource provisioning to satisfy service level agreements and handle power consumption to avoid the greenhouse effect. Additionally, the performance evaluation of the proposed method is performed using Bitbrains workload traces. The experimental findings demonstrate that the proposed model achieves better accuracy and lower errors.