Optimizing Massively Parallel Simulations of Infection Spread Through Air-Travel for Policy Analysis Ashok Srinivasan Dept. of Computer Science Florida State University Tallahassee, USA Email: asriniva@cs.fsu.edu C.D. Sudheer IBM Research New Delhi, India Email: sudheer.chunduri@in.ibm.com Sirish Namilae Aerospace Engineering Department Embry-Riddle Aeronautical University, USA Email: namilaes@erau.edu Abstract—Project VIPRA [1] uses a new approach to modeling the potential spread of infections in airplanes, which involves tracking detailed movements of individual passengers. Inherent uncertainties are parameterized, and a parameter sweep carried out in this space to identify potential vulnerabilities. Simulation time is a major bottleneck for exploration of ‘what-if’ scenarios in a policy-making context under real-world time constraints. This paper identifies important bottlenecks to efficient computa- tion: inefficiency in workflow, parallel IO, and load imbalance. Our solutions to the above problems include modifying the workflow, optimizing parallel IO, and a new scheme to predict computational time, which leads to efficient load balancing on fewer nodes than currently required. Our techniques reduce the computational time from several hours on 69,000 cores to around 20 minutes on around 39,000 cores on the Blue Waters machine for the same computation. The significance of this paper lies in identifying performance bottlenecks in this class of applications, which is crucial to public health, and presenting a solution that is effective in practice. I. Introduction Air travel has been identified as a leading factor in the spread of several infections [2], [3], [4], [5], [6], [7] and this has motivated calls for limitations on air travel during the current Ebola outbreak. However, such limitations carry considerable economic and human costs. Consequently, it is necessary to evaluate the extent of impact of air travel on spread of Ebola and to also identify policy options that can mitigate its spread without major disruption to air travel. Computer simulations play a crucial role in evaluating viable policy options and exploring potential consequences of decisions taken by policy makers. For simulations to be effective, they need to be able to provide insight into the consequences of different policy choices that decision makers may make, and produce results under the time constraints required for quick action. Unfortunately, conventional models for infection spread through air-travel are too coarse-scaled to suggest fine-tuned policies, because they are typically based on analysis of aggregate passenger data. These models cannot account for changes in passenger interaction patterns due to changes in policies or procedures, which in turn influence infection spread. In addition, this approach typically requires good data, and data are often scant during the initial stages of an infection. Fig. 1. Visualization of the initial state of a simulation Project VIPRA takes a different approach, which involves tracking the trajectory of each individual passenger through the SPED model, which is a fine-scale causal model. (Figure 1 shows the initial state of a simulation of passengers deplaning.) The initial focus of this project is on Ebola, which is spread by contact with fluids of infected persons, either through direct contact or indirectly through contact with a common surface. We have shown that seating arrangement and boarding and disembarkation procedures play a major role in such contact. Inherent uncertainties in human behavior make precise pre- diction of human movement or infection transmission difficult. This is magnified by the practical reality that models for all factors influencing an epidemic are not available, especially in the initial stages. Such sources of uncertainty are param- eterized and our goal is to identify potential vulnerabilities in different policy or procedural choices across the parameter space. The goal is not to generate a single likely prediction but, rather, a set of possible scenarios, and to identify vulner- abilities due to any of these possible scenarios. The large parameter space 1 due to the various sources of 1 We use the term ’parameter category’ for a type of parameter, such as passenger velocity in the absence of other passengers in the vicinity. For each parameter category, we may choose a range of values in order to deal with uncertainty in its exact value. The total number of parameters, which is referred to as ’parameter combinations’ in certain contexts, is the product of the number of parameters for each parameter category. The large parameter space arises from a large number of parameter combinations.