Predicting the Impact of Job Placement on System Performance for z/OS: A Machine Learning Approach

Vivek MISHRA

Master's thesis

Abstract:

Efficient utilization of large-scale HPC systems, including z/OS, prioritizes resource management and job scheduling. Modern job scheduling systems need estimates of total wall time and CPU time at job submission to make reliable scheduling decisions. However, z/OS lacks a method for specifying these estimates at submission. This thesis investigates the impact of different job metadata on wall and CPU time prediction using simple conditional-driven programming, machine learning methods. Results indicate that for z/OS, complex machine learning models perform as well as simple condition-driven programming based on job names.

Abstract:

Efficient utilization of large-scale HPC systems, including z/OS, prioritizes resource management and job scheduling. Modern job scheduling systems need estimates of total wall time and CPU time at job submission to make reliable scheduling decisions. However, z/OS lacks a method for specifying these estimates at submission. This thesis investigates the impact of different job metadata on wall and CPU time prediction using simple conditional-driven programming, machine learning methods. Results indicate that for z/OS, complex machine learning models perform as well as simple condition-driven programming based on job names.

Keywords

z/OS, Machine Learning, ML, Artificial Intelligence, AI, IBM, Prediction, Resource, Wall Time, CPU Time, Execution Time, Elapsed Time, LSTM, ANN, Neural Networks, Decision Trees, XGBoost, Lookup Table, HPC, systems, Job scheduling

Language used: English

Date on which the thesis was submitted / produced: 20. 8. 2024

Thesis defence

• Supervisor: prof. Dr. Andreas Kassler

UNIVERSITY OF SOUTH BOHEMIA IN ČESKÉ BUDĚJOVICE

Faculty of Science

Master programme / field:
Artificial Intelligence and Data Science / Artificial Intelligence and Data Science