Optimised Task Placement for MLOps

To accelerate Machine Learning Operations (MLOps) workflows, where Machine Learning (ML) pipelines are the most time-consuming element, heterogeneous environments can be used. However, to take full advantage of this heterogeneity, it is essential to determine the appropriate machine for each task in the pipeline. In this paper, we present a task placement system for ML pipelines that automatically assigns tasks to the most appropriate machine, with the goal of reducing the overall execution time. The placement process is divided into two phases: pipeline scheduling and task placement. The scheduling phase uses the Shortest Job First (SJF) algorithm to determine the pipeline execution order, while the task placement phase uses a heuristic-based approach. Results show that the proposed solution significantly outperforms baseline strategies, achieving reductions in total execution time and average waiting time.