Selection of computational environments for PSP processing on scientific gateways.

Area of Science:

• Computational Biology
• Bioinformatics
• Scientific Computing

Background:

• Science Gateways are crucial for academic research, offering flexibility and ease of use.
• Performance bottlenecks in Science Gateways can lead to wasted resources and suboptimal scientific results.
• Failures within specific systems, such as Galaxy, necessitate detailed investigation.

Purpose of the Study:

• To investigate performance failures in a Galaxy system, specifically within protein structure prediction experiments.
• To propose a novel component for Science Gateways to define the relationship between general parameters and machine capacity.
• To optimize machine setup in heterogeneous environments using machine learning strategies.

Main Methods:

• Analysis of failures within a Galaxy system during protein structure prediction.
• Development of a novel Science Gateway component for parameter-capacity relationship definition.
• Application of machine-learning strategies, including Support Vector Regression (SVR), for optimal machine setup.
• Utilizing historical datasets to train the SVR model.

Main Results:

• A comprehensive overview of the Galaxy platform's organization and workload behavior was achieved.
• The study demonstrated the effectiveness of machine-learning strategies in identifying optimal machine configurations.
• A Support Vector Regression model, trained on historical data, proved to be an effective learning module.
• Varied platform configurations were shown to be valuable infrastructure components for Science Gateways.

Conclusions:

• Investing in local cluster infrastructures is advantageous for scientific experiments within Science Gateways.
• Optimized machine setups enhance resource utilization and improve the quality of scientific results.
• The proposed Science Gateway component and machine-learning approach offer a viable solution for performance optimization.