Abstract
To analyze inner-enterprise cloud cluster performance, the role of workload analysis is of paramount interest to system designers. However, the ever-evolving nature of inner-enterprise cloud platforms such as diversity and spatio-temporal nature of workloads makes evolution diagnosing a challenging task. In this paper, we propose MuCoTrAna-Inner, an evolution diagnosing approach for a large-scale cloud data center based on comparative spatio-temporal trace analysis. Moreover, we present a case study on two representative big traces: Alibaba 2017 trace, and Alibaba 2018 trace. Novel quantitative findings along with the performance bottle-neck inferences and recommendations based on workload analysis are provided. Our multifaceted analyses of the traces and new findings not only reveal interesting insights that are of interest to system designers and administrators, but also establish a new view to diagnosing the evolution of inner-enterprise cloud cluster based on trace analysis.