JobPruner: A machine learning assistant for exploring parameter spaces in HPC applications

Abstract

Abstract-High Performance Computing (HPC) applications are essential for scientists and engineers to create and understand models and their properties. These professionals depend on the execution of large sets of computational jobs that explore combinations of parameter values. Avoiding the execution of unnecessary jobs brings not only speed to these experiments, but also reductions in infrastructure usageparticularly important due to the shift of these applications to HPC cloud platforms. Our hypothesis is that data generated by these experiments can help users in identifying such jobs. To address this hypothesis we need to understand the similarity levels among multiple experiments necessary for job elimination decisions and the steps required to automate this process. In this paper we present a study and a machine learning-based tool called JobPruner to support parameter exploration in HPC experiments. The tool was evaluated with three real-world use cases from different domains including seismic analysis and agronomy. We observed the tool reduced 93% of jobs in a single experiment, while improving quality in most scenarios. In addition, reduction in job executions was possible even considering past experiments with low correlations.

Exploring the Role of Machine Learning in Scientific Workflows: Opportunities and Challenges

“…Therefore, the optimization process for these parametric experiments are considered in various aspects. For example, a subset of values can be selected and explored [86]. In this regard, selecting a subset of values and optimizing the parameters can be observed as a mathematical problem and solved by employing ML techniques.…”

“…In this regard, selecting a subset of values and optimizing the parameters can be observed as a mathematical problem and solved by employing ML techniques. Using learning strategies, the scheduler can find the effective parameters and their optimized values to reduce the search space by comparing the past results and identifying unnecessary jobs, thus speeding up the experiments [86]. Also, real-time interactive control by which the user can modify the progress of execution, change its parameters and the results through analysis visualization, has been reflected in the literature [87].…”

A Search Space Exploration Framework for e-Science Applications

“…These tools mainly focus on managing users' jobs looking into the computational infrastructure. The main difference between our tool and the previous ones is that ours provides useful plugins to accelerate search space explorations [Silva et al 2018]. For instance, Copper presents JobPruner which is a tool that looks into the user's workload and finds patterns from past experiments, which allow users to drastically reduce their search spaces when performing experiments of similar nature.…”

“…These jobs are sent through a message queue for local or remote processing. Copper can utilize techniques like uncertainty quantification, sensitivity analysis, and search space pruning available as plugins to achieve the user's goal [Silva et al 2018]. The subsystem that consumes jobs from the message queue, executes user's software, and returns a result to Copper consists of two main components:…”

“…In this paper, we present a brief overview of Copper and describe how to connect Copper to user applications. The proposed tool has been used to explore search spaces in e-Science applications from different areas such as agriculture, oil & gas, flood simulations, and bioinformatics [Silva et al 2018, Silva et al 2016.…”