Google Cloud Trace: Characterization of Terminated Jobs

Abstract

It is essential to comprehend, analyze and characterize the features of workloads in large-scale clusters to increase system throughput, resource utilization, and resource management. There is a bit of ambiguity about the characteristics of cloud load in data centers, and there has not been much information regarding these workloads that reflect the practical computing requirements of cloud users. For evaluating the effectiveness of recent algorithms in resource allocation and scheduling, developers and organizations have shared their datasets despite difficulties in protecting the confidentiality of the data and the cloud server provider's policies. In this paper, we have extensively evaluated the Google dataset for a period of month. The behavior of the workloads, specifically the job level, and the frequency and pattern of terminated events are described effectively for picturing characteristics of a typical cloud environment.