Purdue University

Project #1: Streamlined Software: Automating user-requested software deployment on Anvil using agile technologies

Simplifying Software Deployment: Empowering users to install complex scientific applications on Anvil with the click of a button! Scientific applications often come with dozens or even hundreds of dependencies with complex build instructions. For example, large software like Tensorflow or Pytorch have hundreds of dependencies and are non-trivial to build from source. The intent of this project is to craft a CI/CD pipeline on Anvil to facilitate scientific software deployment using the Spack package manager. The REU will work on creating web-based tools to automate Spack configuration, software deployment, and testing. As part of this project, we will explore and extend the E4S framework developed by the Exascale Computing Project (ECP). Imagine the possibilities: creating a system that not only eases the burden on staff but also unleashes the full potential of user-driven software deployment.

Requirements: A basic understanding or familiarity with Linux command line tools and/or experience of shell scripting required.

Past experience with Python and compiling open-source software is a plus, but not strictly necessary if candidates can demonstrate successful completion of relevant trainings prior to the start of the REU program:

1. Codecademy - Learn Python at your own pace, on your own schedule and
2. Compiling open-source software from source code

Project #2: Unlocking the Impact of Data: The Power of the Dashboard

Empowering Users: Breaking Down Barriers to Data with an Intuitive Dashboard! Imagine effortlessly tapping into powerful metrics that can help you understand how you are using your computational resources; or where you can wring out better performance without any coding or command-line confusion. In this project, you will break down barriers to data, making complex data more accessible, and transforming the way we handle user resource allocation and code performance information on Anvil! Interested in data analysis, creating user-friendly dashboards, and empowering users to make sound data driven decisions? Then this is the project for you.

Requirements: Basic understanding of Linux system, commands and/or file systems, basic understanding/familiarity with at least one programming language (e.g. Python, C++, Ruby, etc...), and basic understanding/familiarity/experience with data presentation and interpretation.

Past experience with the following skills are a plus, but not strictly necessary: Interactive application/widget development.

Students must be able to complete the following trainings prior to the start of the REU program (trainings are all accessible through RCAC website):

1. Unix 101 and Unix 202 trainings,
2. Anvil 101 training, and
3. (Optional) Anvil Open OnDemand 101 training

Project #3: Gromacs Gateway: Creating User-friendly Molecular Simulations Online

Dive into the world of molecular dynamics with us as we tackle the challenges that await creating easy simulation solutions. Gromacs, the powerhouse for molecular simulations, often feels like a maze with its reliance on Linux commands and a missing link for visualizing structures. This project sets an eye toward integrating Gromacs seamlessly with the Anvil OOD portal. No more grappling with Linux commands or hunting for external tools. This projects goal is to start the process of creating a one-stop-shop for Anvil users Gromacs needs. Perfect for students looking to make waves in computational chemistry through HPC!

Requirements: Familiar with basic Linux commands and environments and a basic understanding of, or ability to, read programing languages like bash or ruby is required.

Past experience with the following skills are a plus, but not strictly necessary if candidate can demonstrate successful completion of a relevant tutorial prior to the start of the REU program:

1. Gromacs simulation tutorial or other computational chemistry simulation training, and
2. Web development tutorial

Project #4: AI-Powered Operational Data Analytics: Enhancing User Experience on Anvil

Picture this: A shared HPC cluster where everyone's racing for computing resources. Your job's in the queue, but the wait time feels like forever. Frustrating, right? This project is diving into the world of operational data analytics, using AI and machine learning to create a smart model to predict job queuing times on Anvil. Our mission? Predicting how long a user will be in the queue! This project is not just crunching numbers but turning messy data into a replicable and explainable model for all users. Once the model is created the project will focus on building a command line tool and integrating it into Anvil's job submission functions so users can automatically see wait time predictions from the model upon job submission, allowing them to plan for when their job will kick off.

Requirements Candidate must be familiar with foundational data analysis methods and frameworks, including statistical analysis, data cleaning, and manipulating data using standard libraries like Pandas.

Past experience with machine learning is a plus, but not strictly necessary if candidate can demonstrate successful completion of a relevant class like Coursera’s Applied Machine Learning in Python course prior to the start of the REU program.

While benchmarking compute and storage performance, students:

• Learned concepts surrounding benchmarking of HPC systems, including HPL, HPCG, STREAM, and IO500
• Measured performance of Anvil's compute nodes, GPU nodes, scratch, project, and BeeOND filesystems
• Established baselines for system performance for a continuous measurement framework

To improve data collection and reporting, students:

• Configured multiple modules on Anvil's XDMoD instance and improved data collection and reporting of system metrics
• PCP, SUPREMM
• Open OnDemand data collection
• Continuous measurement framework via Application Kernels

For system environmental measurements, students:

• Created monitoring of Anvil's power consumption at the node and rack levels using Prometheus and Grafana

Enhance Anvil's cybersecurity posture, utilizing an NSF CICI-funded intrusion detection system to monitor aspects of Anvil's network and enable visualization-driven insights into network traffic and cybersecurity alerts. Students helped with:

• Visualization of network traffic trends on Anvil and Purdue networks
• Dashboard development based on Zeek IDS protocol logs (TCD, UDP, SSH, HTTP)
• Port scanning
• Per-host traffic visualization