CAMPAIGN: an open-source library of GPU-accelerated data clustering algorithms.

Area of Science:

• Bioinformatics
• High-Performance Computing
• Data Science

Background:

• Data clustering is crucial for data analysis, especially in computationally intensive bioinformatics applications with large datasets.
• Sequential clustering algorithms struggle to meet the performance demands of modern big data challenges.
• Massively parallel processing architectures are essential for efficient analysis of large-scale biological data.

Purpose of the Study:

• To introduce CAMPAIGN (Clustering Algorithms for Massively Parallel Architectures, Including GPU Nodes), a specialized library for data clustering.
• To provide a central resource for data clustering algorithms optimized for massively parallel architectures, particularly GPUs.
• To accelerate bioinformatics data analysis through efficient, parallelized clustering techniques.

Main Methods:

• Developed CAMPAIGN using C for CUDA, specifically targeting Nvidia GPUs.
• Implemented a suite of data clustering algorithms designed for parallel execution.
• Structured the library for extensibility to future platforms like OpenCL.

Main Results:

• CAMPAIGN achieves up to a two-orders-of-magnitude speed-up compared to traditional CPU-based clustering algorithms.
• The library is released as an open-source resource under the LGPL license.
• CAMPAIGN is designed for easy extension and community contributions.

Conclusions:

• CAMPAIGN significantly enhances the performance of data clustering in bioinformatics.
• The GPU-accelerated library provides a scalable and efficient solution for large-scale data analysis.
• CAMPAIGN serves as a valuable, extensible open-source tool for the scientific community.