Elliott Vetle Forde
MIT EECS | Nutanix Undergraduate Research and Innovation Scholar
Automatically Exploiting Data Locality in Speculative Parallel Code
2017–2018
Electrical Engineering and Computer Science
- Systems and Networking
Daniel Sanchez
Swarm is a novel task execution model, task management scheme, conflict-detection scheme, and distributed commit protocol that allows for large-scale utilization of ordered parallelism. During simulation, serial applications broken down into tasks and rewritten in Swarm may see speedups of up to hundreds of times. To improve this already massive performance boost, the task management scheme can be improved to be courteous to the data each fine-grained task accesses allowing locality to be exploited and thus reducing data transfer overhead and the likelihood of aborts.
After hearing about SuperUROP I figured it would be a great way to get my hands dirty with legitimate research. I will be working in computation architecture a field very different from my previous software experiences. Planning on pursuing an MEng at MIT I find it extremely valuable to have the opportunity to explore this field with the safety of a mentor and the SuperUROP program to guide me through.