Interested in GSoC 2026 Project 3: Benchmarking and performance optimization

[Amarendra22]

Hello,

I am Amarendra, a B.Tech undergrad at IIT Kharagpur. I have a strong experience working in Python, machine learning and structured software development. I have worked extensively with data-intensive workflows, performance-sensitive code and modular project design, and I am particularly interested in contributing to MDAnalysis for GSoC 2026 — specifically to Project 3: Benchmarking and Performance Optimization.

I am especially drawn to this project because performance engineering and systematic evaluation of code quality are areas I genuinely enjoy. Through my experience on working with data-intensive pipelines and ML systems, I have gained experience with profiling tools, identifying bottlenecks, improving efficiency and writing maintainable code. I appreciate how structured benchmarking helps ensure long-term scalability and stability in scientific libraries. Over the past semester, I had worked with Slack’s Model optimization using ML for my Bachelor Thesis and I have also published a first-author research paper titled “Spiking Neural Network for Cross-Market Portfolio Optimization in Financial Markets: A Neuromorphic Computing approach.”

From reading the project description and roadmap, I understand that the goal is to expand ASV benchmark coverage across major core functionalities, analyse performance trends and prioritize optimization targets. I find this particularly impactful because improving performance directly benefits the broader scientific community relying on MDAnalysis.

I am currently working on:
• Exploring the MDAnalysis codebase and understanding core modules
• Reviewing the existing ASV benchmarking setup
• Studying profiling tools such as cProfile and line_profiler
• Going through related performance issues and discussions

I would begin with small benchmark contributions and gradually expand coverage while understanding the library’s performance-critical paths. I would also appreciate guidance on which core areas currently need benchmark coverage most urgently.
Looking forward to engaging with the community and contributing meaningfully.

Best regards,
Amarendra

[orbeckst]

For right now, if you're interested in performance, you should familiarize yourself with ASV and perhaps write a simple benchmark — just choose anything that's not covered already. It's more important to learn the process than to be exhaustive.

The MDAnalysis.lib module has some code that is used repeatedly. All coordinate readers/writers are important. Many analysis tools are not covered and are good target.

[Amarendra22]

Thank you for your reply. Currently, I am focusing on benchmarks whose execution time is greater than 5 ms, as these are more likely to reveal meaningful performance bottlenecks. My plan is to analyze these benchmarks, identify the corresponding sections of the codebase, and make potential optimization changes.

[Amarendra22]

Hi @orbeckst ,

I started by adding ASV benchmarks for several performance-critical functions in MDAnalysis.lib, since these kernels are used repeatedly across many analysis modules.

So far I have implemented benchmarks for:

• distance_array
• self_distance_array
• calc_dihedrals
• FastNS
• NeighborSearch
• Contacts Analysis

These benchmarks are parameterized by the number of atoms so that we can observe scaling behavior and detect performance regressions more clearly.

PR: #5291 and #5292

Running the benchmarks locally through ASV works correctly and generates results for different Python versions.

Next, I plan to:

• continue expanding benchmark coverage in MDAnalysis.lib
• add benchmarks for additional analysis modules that are not yet covered
• start profiling some of these functions to identify potential optimization opportunities.

If there are particular areas (modules or functions) where additional benchmark coverage would be especially useful, I would be happy to focus there.

Best Regards,
Amarendra.