Acquisition of a High-Performance Computing System for Interdisciplinary Research and Teaching

Amy Sliva (Principal Investigator)

amysliva@kings.edu

Avik Mahata (Co-Principal Investigator)

Malitha C Rajapaksha Manikkunambi (Co-Principal Investigator)

NSF Award # 2407535

Project Overview

• Objective: Acquire an HPC system to support interdisciplinary research and teaching

​

• Location: McGowan School of Business, King’s College

​

• Training and Education: Mechanical and Civil Engineering (ENGR 300A/ENGR 300LA, ME410), Computer Science (CS 380, CS 490) Mathematics, Physics and Chemistry (MATH 365, PHYS 496/497, CHEM 197)

​

• Key Research Areas: Materials Science, Computer Science, Civil Engineering, Mathematics

​

​

Machine Overview

• Compute CPU/GPU Nodes

​

16 General Compute nodes with 32 CPUs each capable of creating 2 threads or 128 simultaneous simulations (2x Intel 32-Core Xeon Gold )

​

2 GPU Servers with 2 GPUs each (NVIDIA Quadro RTX A6000)

​

Storage

​

• 384TB Usable data storage

​

Network System

​

• NDR InfiniBand Networking

Research Activities Enabled

1. Functional 2D High entropy Materials (Mechanical Engineering & Materials Science)

​

• Machine Learning & AI for sociocultural analysis (Computer Science)

​

• Multi-physics simulations for pavement research (Civil Engineering)

Functional High Entropy Materials

Functional High Entropy Materials

High Entropy Alloy Nanoparticle

High Entropy Alloy Nanoparticle

Fig. 8. High Entropy Alloy Research Workflow using both Computational and Experimental data, integrating state-of-the-art novel AI/ML algorithms

Modeling Sociocultural Systems

• Desire in many domains to better take advantage of stochastic models for decision support
• Leverage data sources and analytics otherwise intractable for human decision making
• Estimate the behavior of a system under varying conditions
• Explore the possible outcomes of decisions
• Sociocultural system
• Behavioral dynamics caused by a confluence of interacting and interdependent social, cultural, economic, and political systems
• Representing complex, stochastic relationships and influences
• Probabilistic graphical models, Markov logic networks, probabilistic soft logic, timed influence networks

​

Timed Influence Networks

• Timed influence networks (TIN)
• Probabilistic graphical model
• Tripartite graph—random variables, decision variables, outcome variables
• Transforms into a Bayesian network for inference and analysis
• Inference
• Compute posterior probabilities for nodes in the network
• Course of Action (COA): For a model with decision nodes d₁,…,d_k​�COA = Δ(δ₁,…,δ_k) specifying a value for each d_i
• Goal: find the optimal course of action (COA)
• Values of decision variables that maximize/minimize the probability of the outcome variable o
• P*(o) = max_Δ P_Δ(o)
• Δ* =  argmax_Δ E_Δ(o)
• Incorporate observed evidence or “what if?” scenarios by fixing values of some random variables
• Full powerset of variables computationally intractable
• Approximate interaction effect with Monte Carlo simulation over the model
• Requires large-scale computation resources for even modestly sized models of real-world decisions

​

Case Study: Food Security in Gambella, Ethiopia 2019

• Illustrate the complexity and stochastic nature of sociocultural systems
• Interdependent subsystems
• Dynamic model with seasonal �variations from agricultural cycle
• Several possible decision points
• Data-driven model, simplified� for analysis—plausible, but �not intended for real-world �decision making
• MRI hardware will enable�models of higher fidelity

​

Model Implementation as TIN

Model Experimentation

• Random variables represent
• Agroeconomic processes (e.g., agricultural inputs, crop yields, transport)
• Temporal effects of precipitation (e.g., drought and flood conditions during different parts of the agricultural cycle)
• Sociopolitical events (e.g., civil conflict and human migration)
• Impact of policy interventions to reduce likelihood of famine
• Decision variables
• Increase in food imports (current purchases)
• Provision of Food Aid
• Provision of Direct Aid
• Conflict Resolution
• Ran model inference for 10,000 random samples
• 10,000 samples where variables were set to minimum
• Same 10,000 samples where variables were set to maximum

​

​

​

Primary distresses in the design of flexible pavements

• A Mechanistic-Empirical model developed to predict flexible (asphalt) pavement temperature based on daily cycling for a one-year period to include seasonal effects

​

Objective of current research

• Developing a Mechanistic-Empirical model to predict distresses of flexible (asphalt) pavements using viscoelastic modeling of asphalt and pavement temperature predictions from developed model

Primary distresses used in the design of flexible pavements

Figures from :https://www.pavementinteractive.org/; https://www.tensar.co.uk

Findings - Measured & Predicted Pavement Temperatures at Different Depths

​

Summer

Winter

Research Design - Simplified one-Dimensional Finite Element model

Heat transfer model with Boundary Conditions

Software Platform: COMSOL Multiphysics®

Used 1-D Model

Q_solar = Q_conduction + Q_convection + Q _radiation