Undergrad Research Project - Exploiting Commutativity to Accelerate Parallel Big Data Computations

Spring 2016

Student
Junhong Li
Advisor
Brandon Lucia
Project description

This project is to exploit the improvement of performance, for parallelizable programs, under a new computer architecture design.

This new architecture design focuses on optimizing commutative computations, making private data copies in different parallel threads, eliminating unnecessary, slow access to shared memory.

We will focus on modifying existing parallel programs, to exploit the new architecture, achieving scalable parallel performance improvements. We will especially focus on big data applications that process complex biological networks and other graph structured computations.

We will evaluate the work using an existing simulator that models the commutativity-optimized architecture. Additionally, we will modify the simulator to explore the design space of possible hardware configurations, identifying valuable tradeoffs of complexity, performance, and parallelism.

Return to project list