HPC Benchmarker [2019]
Concept
An easy to use web application for running common C/C++ and Cuda algorithms (currently two supported) on UIC ACER servers in the form of a benchmark test. The C/C++ and Cuda results can then be compared against one another in a visually meaningful way.
This project was done as a research project during my Summer 2019 internship for ACER@UIC. It was found that for parallelizable tasks, Cuda accelerated code is exponentially more efficient than traditional computations completed on the main processor.
Features
As described above, this project was designed to allow any authorized user to easily live run and see the performance differences between C/C++ (CPU) and Cuda (GPU) code. In order to accomplish this, the following features were implemented:
Browser support for ease of use and operating system independence (requires project to be hosted first)
SSH support to gain access to UIC servers (for security purposes, only manually authorized users can login)
Basic two factor authentication support via Duo Mobile push notifications (for increases security purposes)
Footer links for quick help dialog boxes
Automated shell scripts to load, compile, run, and fetch the output of each benchmark on the back-end
One button to run a single benchmark and display its results to the user on the front-end
Benchmarks customizable with different parameters for different results
Benchmark results saved so that previous runs still show
One button to view the graphical results of each benchmark (with data points that can be hovered for information on that specific run)
Click here for a more complete set of documentation
Release
The source code can be viewed on GitHub.