Undergrad Research Project - Selectively Consistent Approximate Parallel Execution

Spring 2017

Student
Devon White
Advisor
Brandon Lucia
Project description

This past year, Brandon Lucia and a small team worked on "Selectively Consistent Approximate Parallel Execution." The goal of the project was to see if selectively disabling coherence on non-critical variables in a multi-processor system would yield performance gains, and how much of a hit to accuracy the performance gains would come with. The speed-up measured in simulation was very significant, with an acceptable loss of accuracy. My project for this semester, therefore, will be an extension of this project. The goal will be to implement the Selectively Consistent Approximate Parallel Execution model in a multi-processor system on an FPGA. This will be done by finding a pre-existing, open source, Verilog implementation of a multi-processor system and modifying it, then synthesizing and putting it onto an FPGA to measure actual performance in hardware. I will compare this performance to that of the original simulation for our results to draw conclusions.

Return to project list