Virginia Bioinformatics Institute, collaborators receive $1.45 million to develop petascale computer modeling capabilities
The National Science Foundation (NSF) has awarded a four-year, $1.45-million grant to the Network Dynamics and Simulation Science Laboratory at the Virginia Bioinformatics Institute at Virginia Tech and partners to develop petascale computing environments that model billions of individuals in extremely large social and information networks.
The anticipated arrival of high-performance computers that routinely perform one quadrillion (one million billion) operations per second means that complex studies of global populations at the level of the individual can realistically be simulated on distributed computer networks.
The goal of the proposal “Coupled Models of Diffusion and Individual Behavior Over Extremely Large Social Networks” is to use new computer technology breakthroughs to study events like disease pandemics, financial crises, as well as the spread of opinions, attitudes, or social beliefs, through populations on a global scale.
Current state-of-the-art agent-based computer models can simulate the spread of a disease like influenza through a population the size of the United States. Petascale modeling would make comparable agent-based studies of disease transmission possible for global populations.
The laboratory will work with partners at the Brookings Institution, Indiana University, Northwestern University, and the University of Illinois at Urbana-Champaign, to develop models and algorithms that support the work of researchers, policy- and decision-makers who want to examine and probe individual, and group behaviors in these extremely large global social networks.
Madhav Marathe, deputy director of the Network Dynamics and Simulation Science Laboratory and professor in the Department of Computer Science at Virginia Tech and principal investigator on the proposal, remarked: “Underpinning this project is a desire to create some of the next-generation computational tools and environments that will be needed to enable future research by social, biological, and computational scientists. We anticipate unprecedented increases in scaling and execution speeds for computer processors in the years ahead.”
These improvements will make it possible to look in parallel at multiple diffusions and behaviors as they evolve and influence different interactions in these extremely large social networks. We hope to be able to resolve these large networks of interactions all the way down to the level of the individual. Representing the coupled and co-evolving aspects of the networks and their constituent elements is a significant computing challenge, one that needs to be met if we are to understand these complex socio-technical phenomena,” Marathe added.
The collaborators will construct a petascale computational modeling environment – MTML-Sim – that will scale to billions of individuals and their social and information networks.
The scaling will be achieved by developing innovative parallel algorithms as well as their implementations that will allow researchers to map the networks on petascale computing environments that are in the process of being built and deployed at places such as the National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign. This environment will be used to simultaneously test multiple theories of social interaction amongst individuals and groups.
Noshir Contractor, the Jane S. and William J. White Professor of Behavioral Sciences in the School of Engineering, School of Communication, and the Kellogg School of Management at Northwestern University and co-principal investigator on the proposal, remarked: “Petascale computer modeling will open up many exciting opportunities to explore global multidimensional social and knowledge networks. It will enable us to theorize, simulate, and empirically validate how these dynamic networks form and evolve. The insights from this research will have unprecedented relevance to the work of policy makers interested in studying and managing a wide range of grand societal challenges.”
Sanjay Kale, professor at the Department of Computer Science, University of Illinois at Urbana-Champaign and co-principal investigator on the proposal, commented: “Our efforts will focus on improving the performance and productivity of agent-based modeling applications on these 100,000+ processor petascale computer architectures. Guided by direct collaboration with application developers, we will make enhancements to the Charm++ runtime system and associated performance analysis tools, which will give us a handle on designing and improving the software environment to accelerate application development for the next generation of petascale computer systems.”
Dimitris Nikolopoulos , associate professor in the Department of Computer Science, College of Engineering, at Virginia Tech, added: “A key part of this project will be the development of new software technology for enabling petascale computational modeling environments on processors with many cores coupled tightly with computational accelerators. We will be working closely with all collaborators to explore how future hardware technologies can catalyze the discovery of next-generation computer modeling solutions.”
Xizhou Feng, senior simulation science software developer at the Virginia Bioinformatics laboratory, commented: “One of our goals in this project is to deliver a high-performance software environment that will work hand-in-hand with the state-of-the-art computer architectures. By combining advances in both systems and software, we hope to achieve the required scalability, usability, and efficiency for modeling a class of highly complex systems that are critical to the study of a wide range of global socio-technical challenges.”
Keith Bisset, senior simulation science system software developer at the lab, concluded: “The transdisciplinary approach we will use for the study of these extremely large networks should greatly enhance the explanatory value of the global models we are interested in and their utility as a platform for policy-based decision making.”
The Network Dynamics and Simulation Science Laboratory pursues an advanced research and development program for interaction-based modeling, simulation, and associated analysis, experimental design, and decision support tools for understanding large biological, information, social, and technological systems. Extremely detailed, multi-scale computer simulations allow theoretical and experimental investigation of these systems.
