ChainLab Grid

Summary

ChainLab Grid's mission is to make computing resources accessible via decentralized crowdfunding of compute power.

Share your task with the network, have people from all-over the world help get the task done while their work is securely validated and aggregated by our Trusted Execution Environments, powered by Oasis.

We offer a welcoming UI for task sharing and assigning, paired with a service that uses your compute power the way you want it and after that it's on us, we do the validation, we do the aggregation and we serve back your results, securely and only for you, back in the UI.

The whole deal

System Overview

Our system is designed to manage tasks efficiently using a modular architecture:

• Front-End (Web UI): This is the user interface where you can submit new tasks, pick-up work, keep an eye on their progress and see your results.

• Main Contract: Think of this as the system’s central hub-it coordinates all task management activities.

• ROFL Components: These handle all the heavy lifting off-chain, including processing, validation and aggregation.

• Subcontracts: Responsible for enlisting and distributing subtasks to different clients.

• Individual Clients: These are the workers that actually execute the subtasks they’re given.

How It Works

1. Task Submission

• You submit a task through the Front-End.

• The Main Contract receives it and alerts the ROFL Request Listener to validate the task.

2. Task Expansion

• Once the task is validated, the Main Contract breaks it down into smaller subtasks.

• These subtasks are then handed off to the appropriate Subcontracts.

3. Subtask Assignment

• The Front-End checks with the Main Contract to see which subtasks are available.

• Subtasks are picked by users based on their current capacity. Anyone can pick anything as long as they have the resources for it.

4. Execution and Validation

• Clients work on their assigned subtasks, they are linked by their wallets, and send the results back through the Subcontracts.

• The ROFL Sub-Task Validator checks each result for accuracy and authenticity.

• Finally, the ROFL Task Validator performs a last round of validation for the whole Task.

5. Aggregation and Completion

• The ROFL Task Aggregator gathers all the validated results and performs the final operation.

• These results are encrypted and sent back to the Main Contract for final processing.

ROFL Integration

All ROFL components are designed to run inside Trusted Execution Environments (TEEs), which means that off-chain computations are both secure and verifiable. These components communicate with on-chain contracts through secure, authenticated channels, so your data stays safe and confidential throughout the process.

Key Architectural Features

• Modularity: Each component has a clear role, making the system easy to manage and extend.

• Asynchronous Processing: Tasks and validations happen independently, which improves efficiency.

• Security: TEEs ensure that off-chain operations are protected.

• Scalability: The modular design allows the system to handle a large number of tasks without breaking a sweat.

Practical Considerations

Right now, our PyTorch-based validators are pretty resource-intensive. For demonstration purposes, we’re running everything locally using unsecured contracts. These validators are a key part of our ROFL framework and are meant to run inside TEEs for maximum security and verifiability. However, the current test pod infrastructure (just dual-core CPUs and 4GB of RAM) isn’t quite up to the task-PyTorch typically needs at least 8GB of RAM and more powerful processors for smooth operation.

Despite these limitations, we’ve already implemented all the logic needed for seamless ROFL integration, including generic methods for calling secure contract functions. So, once we upgrade our hardware, moving to a secure, production-ready environment should be a smooth and straightforward process.

Presentation

Presentation Link

Github Organization

Github Link

Discord Usernames

• ionel_patrascu
• bogdann007
• baciugeorge
• pcv0963
• askingalexander