An inter-cell resource usage analysis of large-scale datacentre trace logs

In recent years, modern cluster management systems are facing large volumes of workloads with complex queries. Smart processing of such workloads is essential, which requires a comprehensive understanding of large-scare data centres and the way workloads are processed in them. Understanding of workload and datacentre characteristics are also expected to contribute to the development of prediction models. The recent publication of the 2019 Google trace logs is expected to provide useful insights to researchers of energy efficient datacentres. However, insights from this trace log still remains largely uncovered. This paper presents a comprehensive analysis on the distribution of machine resource utilisation across various cells encompassed in the 2019 Google cluster trace log, uncovering various essential workload behavioural insights such as execution duration, workload task composition, CPU/memory utilisation etc. along with a longitudinal comparative analysis with two other large datasets, namely the 2018 Alibaba and 2011 Google trace logs.