Modified Critical Path Task Scheduler

This project implements the Modified Critical Path (MCP) algorithm for task scheduling under constraints (dependencies, memory, heterogeneous resources).
Developed as part of a CentraleSupélec project in collaboration with Aneo (consulting in HPC, Cloud, AI/ML).

Features

• Implementation of the MCP scheduling algorithm based on critical path theory
• Dynamic allocation of heterogeneous resources (memory constraints, failures, elasticity)
• Consideration of inter-processor communication costs
• Scalable to graphs with up to 100,000 nodes
• Deployment on AWS Cloud using S3 and Lambda for high-performance simulations

Results

• Efficient scheduling close to the theoretical critical path
• Robust handling of failures and dynamic resource availability
• Successful scaling to large graphs with cloud deployment

Authors

• Hugo Vuacheux – CentraleSupélec
• Deodato V. Bastos Neto – CentraleSupélec

Supervised by Aneo (HPC & Cloud consulting).