Characterization of the behavior of I / O Using Unsupervised learning

• Pablo Pavan UFRGS
• Jean Luca Bez UFRGS
• Matheus Serpa UFRGS
• Francieli Zanon Boito INRIA
• Philippe Olivier Alexandre Navaux UFRGS

DOI: https://doi.org/10.5753/eradrs.2020.10787

Abstract

In applications HPC operations of I / O bottlenecks are due to the difference between processing speed and a given access. Thus, characterizing the operations can assist in the search for performance. Thus, this work proposes an unsupervised learning approach to characterize I / O. Using data Intrepid Supercomputer, we were able to identify the main feature of its applications.

Keywords: Evaluation, Performance Measurement and Prediction, File systems and input and high output performance, Clusters (clusters)

References

Calzarossa, M. C., Massari, L., and Tessera, D. (2016). Workload characterization: A survey revisited. ACM Computing Surveys (CSUR), 48(3):48.

Di, S., Kondo, D., and Cappello, F. (2014). Characterizing and modeling cloud applications/ jobs on a google data center. The Journal of Supercomputing, 69(1):139–160.

Liu, Y., Gunasekaran, R., Ma, X., and Vazhkudai, S. S. (2016). Server-side log data analytics for i/o workload characterization and coordination on large shared storage systems. In SC’16: International Conference for High Performance Computing, pages 819–829. IEEE.

Zoll, Q., Zhu, Y., and Feng, D. (2010). A study of self-similarity in parallel I/O workloads. In Mass Storage Systems and Technologies (MSST), 2010 IEEE 26th Symposium on, pages 1–6. IEEE.