Analysis and Classification of Job Multiple Characteristics on Supercomputers

The diversity and complexity of applications grow, with the development of HPC systems. In-depth investigation and study of the job characteristics are critical for system optimization, management and application optimization. Our study in the paper strives to explore some important characteristics based on long-term historical data. We get many useful insights and findings about runtime distribution, job status and I/O behaviors. Besides, we present a grouping method to classify the jobs into different groups. Evaluations about the runtime and I/O prove that jobs with similar names have similar characteristics.