Implementation of Parallel Computation for Fast Optimization with the Genetic Algorithm in Numerical Electromagnetics
Primary Faculty Mentor’s Name
Dr. Sungkyun Lim
Proposal Track
Student
Session Format
Paper Presentation
Abstract
An advanced system of parallel computation was developed. The system is comprised of multiple software and computers that assist in processing data, which act as cloud computing.
The system was divided into 3 parts. The first is the data, which is the information stored by the host computer that needs to be processed by a series of calculations. For this project, the software used to create and process the data is Mentor Graphics IE3D, a numerical electromagnetics software. Any simulation software can be used with a few modifications. Second is the software used run the entire system. The software is MATLAB, and it serves as a connector between all of the independent data. MATLAB also, in this case, chooses the variables of the data using the Genetic algorithm, which is a system of equations that choses optimum values of a certain situation. Last is the actual hardware, which are the computers and their network. The computers are all already interconnected by a network provided by the institution. Each one has a total of 24 cores and 32 GB of RAM, to allow the maximum of simultaneous simulations.
The system was tested as followed: MATLAB uses its inbuilt app named Genetic Algorithm (GA) with user defined variables to create the data to be simulated. The data is then simulated by I3ED, which only allows one simulation per computer, regardless of the number of cores or processors. In order to solve this problem and reduce the number of computers, multiple Virtual Machines (VMs from VMware) are implemented on one computer, which act as independent computers. MATLAB is given a small portion of the code unique to each simulation software that commands all of the computers and VMs to run the data. It then returns all of the data to the GA which in turn creates new variables based on the previous results.
The new parallel processing system was then compared to the original series computation. It was shown that as the number of parallel simulations increased, the start to finish ratio of the parallel computation and series also grew. This simulation has twenty different simultaneous simulations however, the parallel computing was closer to 18 times shorter instead of twenty, using twenty copies of the exact same data to ensure there were no outside variables.
There is no limit to the number of simultaneous simulations, provided that there are sufficient hardware to do the computations.
Keywords
Genetic Algorithm, Parallel Computation, MATLAB, IE3D, Virtual Machines, Electromagnetics
Award Consideration
1
Location
Room 2911
Presentation Year
2015
Start Date
11-7-2015 9:00 AM
End Date
11-7-2015 10:00 AM
Publication Type and Release Option
Presentation (Open Access)
Recommended Citation
Lucien, Deon J., "Implementation of Parallel Computation for Fast Optimization with the Genetic Algorithm in Numerical Electromagnetics" (2015). Georgia Undergraduate Research Conference (2014-2015). 52.
https://digitalcommons.georgiasouthern.edu/gurc/2015/2015/52
Implementation of Parallel Computation for Fast Optimization with the Genetic Algorithm in Numerical Electromagnetics
Room 2911
An advanced system of parallel computation was developed. The system is comprised of multiple software and computers that assist in processing data, which act as cloud computing.
The system was divided into 3 parts. The first is the data, which is the information stored by the host computer that needs to be processed by a series of calculations. For this project, the software used to create and process the data is Mentor Graphics IE3D, a numerical electromagnetics software. Any simulation software can be used with a few modifications. Second is the software used run the entire system. The software is MATLAB, and it serves as a connector between all of the independent data. MATLAB also, in this case, chooses the variables of the data using the Genetic algorithm, which is a system of equations that choses optimum values of a certain situation. Last is the actual hardware, which are the computers and their network. The computers are all already interconnected by a network provided by the institution. Each one has a total of 24 cores and 32 GB of RAM, to allow the maximum of simultaneous simulations.
The system was tested as followed: MATLAB uses its inbuilt app named Genetic Algorithm (GA) with user defined variables to create the data to be simulated. The data is then simulated by I3ED, which only allows one simulation per computer, regardless of the number of cores or processors. In order to solve this problem and reduce the number of computers, multiple Virtual Machines (VMs from VMware) are implemented on one computer, which act as independent computers. MATLAB is given a small portion of the code unique to each simulation software that commands all of the computers and VMs to run the data. It then returns all of the data to the GA which in turn creates new variables based on the previous results.
The new parallel processing system was then compared to the original series computation. It was shown that as the number of parallel simulations increased, the start to finish ratio of the parallel computation and series also grew. This simulation has twenty different simultaneous simulations however, the parallel computing was closer to 18 times shorter instead of twenty, using twenty copies of the exact same data to ensure there were no outside variables.
There is no limit to the number of simultaneous simulations, provided that there are sufficient hardware to do the computations.
