CampusGrid

Project overview

building a supercomputer from idle computers for heavy tasks which take a long time to run on one computer

Inspiration

Built because running ML models on a single laptop was painfully slow, while powerful machines around campus sat idle.

What it does

It lets users share unused computing power across campus so heavy tasks like ML training or simulations run faster.

How we built it

It uses a lightweight server to coordinate peers and WebSockets for communication, with Python handling workload distribution and execution.

Individual contributions

arijit built the backend architecture, including peer registration, task handling, and communication logic using FastAPI and WebSockets and tushar was runniung tests simulataneously so that system remains robust under all conditions

Challenges

Getting reliable peer-to-peer communication working and ensuring tasks run correctly across multiple unpredictable devices.

Accomplishments

Successfully built a working peer-to-peer compute network and proved tasks could run faster using shared idle hardware.

Learnings

Learned how to build and coordinate distributed systems, handle real-time communication, and deal with unpredictable network behavior.

Next steps

Planning to improve scheduling, add authentication, and make the system scalable enough to run across an entire campus network.

Links