Title of Poster / Presentation

Performance Analysis of DroughtHPC

Start Date

10-5-2017 11:00 AM

End Date

10-5-2017 1:00 PM

Subjects

Droughts -- Prediction -- Data processing, Droughts -- Prediction -- Computer simulation, Parallel programs (Computer programs)

Description

We present our performance analysis of DroughtHPC, a software application being developed by an interdisciplinary effort lead by Dr.Moradkhani in the Civil Engineering department, Dr. Daescu in the Math Department, and Dr. Karavanic in the Computer Science Department.

The DroughtHPC application is used to predict drought conditions for a target geographical area. The data used in the prediction are soil conditions, vegetation layers, canopy cover, snow accumulation information from satellites, and meteorological data. DroughtHPC is written in Python and uses two hydrologic models, PRMS [1] and VIC [2], to simulate soil moisture levels. A larger geographical area such as the Columbia river basin, is divided into cells, and a prediction is computed for each cell independently.

Our goal is to identify run-time performance bottlenecks to allow us scale the application to simulate a larger area, such as the continental United States. We have evaluated the factors that affect the application performance such as computation time, time spent on saving data in files, interaction between different phases of the application using these files, and interference from other jobs running on the HPC cluster. We plan to apply the knowledge to obtain a holistic view of the application, and guide us towards effective resource utilization.

Persistent Identifier

http://archives.pdx.edu/ds/psu/20086

Share

COinS
 
May 10th, 11:00 AM May 10th, 1:00 PM

Performance Analysis of DroughtHPC

We present our performance analysis of DroughtHPC, a software application being developed by an interdisciplinary effort lead by Dr.Moradkhani in the Civil Engineering department, Dr. Daescu in the Math Department, and Dr. Karavanic in the Computer Science Department.

The DroughtHPC application is used to predict drought conditions for a target geographical area. The data used in the prediction are soil conditions, vegetation layers, canopy cover, snow accumulation information from satellites, and meteorological data. DroughtHPC is written in Python and uses two hydrologic models, PRMS [1] and VIC [2], to simulate soil moisture levels. A larger geographical area such as the Columbia river basin, is divided into cells, and a prediction is computed for each cell independently.

Our goal is to identify run-time performance bottlenecks to allow us scale the application to simulate a larger area, such as the continental United States. We have evaluated the factors that affect the application performance such as computation time, time spent on saving data in files, interaction between different phases of the application using these files, and interference from other jobs running on the HPC cluster. We plan to apply the knowledge to obtain a holistic view of the application, and guide us towards effective resource utilization.