Integrating Clustering and Learning for Improved Workload Prediction in the Cloud

Good resource management is very important in the cloud and workload prediction is a crucial step towards achieving good resource management. While it is possible to predict the workloads of long-running tasks based on the seasonality in their historical workloads, it is difficult to do so for tasks which do not have such recurring workload patterns. In this paper, we consider a different solution for task workload prediction. Instead of using the historical workload of a task to predict the future workload of the same task, we use the knowledge about the workloads of a pool of tasks to help predict the workloads of new tasks. In this paper, we develop a clustering and learning based approach to realize this concept. First, the workloads of existing tasks are grouped into multiple clusters. Then, neural network is used to learn the characteristics of the workloads of each cluster. For each new task, we collect its initial workload, determine its cluster, and use the trained neural network of its cluster to predict its future workload. Our approach is experimentally evaluated using Google dataset. The results confirm the effectiveness of our integrated scheme.