šŸ‰ Implementation of a Dragonfly+ Topology in the SST Simulator

The Dragonfly+ topology is an advanced interconnection network topology designed to address the scalability and performance limitations of traditional topologies in large-scale supercomputing environments. It builds on the Dragonfly topology, which is known for its high scalability and efficient routing mechanisms, and introduces enhancements for improving network congestion and fault tolerance in massive systems.

In this thesis, the student will implement the Dragonfly+ topology in the Structural Simulation Toolkit (SST) [1], a widely used simulation framework for large-scale systems. The goal of the project is to model and simulate the Dragonfly+ topology within SST and evaluate its performance under various workloads. The student will focus on understanding the impact of the Dragonfly+ design on key performance metrics such as latency, throughput, and scalability, especially as the system size grows.

The implementation will involve designing the network components, integrating them with SSTā€™s simulation environment, and running experiments to analyze the behavior of the Dragonfly+ topology in different network configurations. Depending on the studentā€™s progress and interests, the project could also include optimizing routing strategies or exploring fault-tolerant mechanisms for Dragonfly+.

Skills required:

  • Knowledge of C/C++

[1] Structural Simulation Toolkit (SST)
[2] Dragonfly+: Low Cost Topology for Scaling Datacenters

Approximate composition: 15% State of the art analysis, 25% Theory/Design, 60% Implementation/Experiments