Neil Harvey

Decision-making for foot-and-mouth disease control: objectives matter

Formal decision-analytic methods can be used to frame disease control problems, the first step of which is to define a clear and specific objective. We demonstrate the imperative of framing clearly-defined management objectives in finding optimal control actions for control of disease outbreaks. We illustrate an analysis that can be applied rapidly at the start of an outbreak when there are multiple stakeholders involved with potentially multiple objectives, and when there are also multiple disease models upon which to compare control actions. The output of our analysis frames subsequent discourse between policy-makers, modellers and other stakeholders, by highlighting areas of discord among different management objectives and also among different models used in the analysis. We illustrate this approach in the context of a hypothetical foot-and-mouth disease (FMD) outbreak in Cumbria, UK using outputs from five rigorously-studied simulation models of FMD spread. We present both relative rankings and relative performance of controls within each model and across a range of objectives. Results illustrate how control actions change across both the base metric used to measure management success and across the statistic used to rank control actions according to said metric. This work represents a first step towards reconciling the extensive modelling work on disease control problems with frameworks for structured decision making.

Predictive analytics using statistical, learning, and ensemble methods to support real-time exploration of discrete event simulations

Discrete event simulations (DES) provide a powerful means for modeling complex systems and analyzing their behavior. DES capture all possible interactions between the entities they manage, which makes them highly expressive but also compute-intensive. These computational requirements often impose limitations on the breadth and/or depth of research that can be conducted with a discrete event simulation.This work describes our approach for leveraging the vast quantity of computing and storage resources available in both private organizations and public clouds to enable real-time exploration of discrete event simulations. Rather than directly targeting simulation execution speeds, we autonomously generate and execute novel scenario variants to explore a representative subset of the simulation parameter space. The corresponding outputs from this process are analyzed and used by our framework to produce models that accurately forecast simulation outcomes in real time, providing interactive feedback and facilitating exploratory research.Our framework distributes the workloads associated with generating and executing scenario variants across a range of commodity hardware, including public and private cloud resources. Once the models have been created, we evaluate their performance and improve prediction accuracy by employing dimensionality reduction techniques and ensemble methods. To make these models highly accessible, we provide a user-friendly interface that allows modelers and epidemiologists to modify simulation parameters and see projected outcomes in real time. Our approach enables fast, accurate forecasts of discrete event simulations.The framework copes with high dimensionality and voluminous datasets.We facilitate simulation execution with cycle scavenging and cloud resources.We create and evaluate several predictive models, including ensemble methods.Our framework is made accessible to end users through an interactive web interface.

On the distributed orchestration of stochastic discrete event simulations

Discrete event simulations are a powerful technique for modeling stochastic systems with multiple components where interactions between these components are governed by the probability distribution functions associated with them. Complex discrete event simulations are often computationally intensive with long completion times. This paper describes our solution to the problem of orchestrating the execution of a stochastic, discrete event simulation where computational hot spots evolve spatially over time. Our performance benchmarks report on our ability to balance computational loads in these settings. Copyright

Autonomous, failure-resilient orchestration of distributed discrete event simulations

Discrete event simulations model the behavior of complex, real-world systems. Simulating a wide range of relevant events and conditions naturally provides a more accurate model, but also increases the computational workload associated with the simulation. To manage these processing requirements in a scalable manner, a discrete event simulation can be distributed across a number of computing resources. However, individual tasks in the simulation are stateful, and therefore require inter-task communication and synchronization to produce an accurate model. This property not only complicates the orchestration of the discrete event simulation in a distributed setting, but also makes providing reliable, fault-tolerant execution a challenge, especially when compared to conventional distributed fault tolerance schemes. In this paper, we propose an autonomous agent that provides fault tolerance functionality for discrete event simulations by predicting state changes in the simulation and adjusting its fault tolerance policy accordingly. This allows the system to avoid negatively impacting overall execution times while preserving reliability guarantees. To underscore the viability of our solution, we provide benchmarks of a production discrete event simulation that can sustain failures while running under the supervision of our fault tolerance framework.

Using Distributed Analytics to Enable Real-Time Exploration of Discrete Event Simulations

Discrete event simulations (DES) provide a powerful means for modeling complex systems and analyzing their behavior. DES capture all possible interactions between the entities they manage, which makes them highly expressive but also compute-intensive. These computational requirements often impose limitations on the breadth and/or depth of research that can be conducted with a discrete event simulation. This work describes our approach for leveraging the vast quantity of computing and storage resources available in both private organizations and public clouds to enable real-time exploration of a discrete event simulation. Rather than considering the execution speed of a single simulation run, we autonomously generate novel scenario variants to explore an entire subset of the simulation parameter space. These workloads are orchestrated in a distributed fashion across a wide range of commodity hardware. The resulting outputs are analyzed to produce models that accurately forecast simulation outcomes in real time, providing interactive feedback and bolstering research possibilities.

Autonomous Orchestration of Distributed Discrete Event Simulations in the Presence of Resource Uncertainty

Discrete event simulations model the behavior of complex, real-world systems. Simulating a wide range of events and conditions provides a more nuanced model, but also increases its computational footprint. To manage these processing requirements in a scalable manner, discrete event simulations can be distributed across multiple computing resources. Orchestrating the simulations in a distributed setting involves coping with resource uncertainty. We consider three key aspects of resource uncertainty: resource failures, heterogeneity, and slowdowns. Each of these aspects is managed autonomously, which involves making accurate predictions of future execution times and latencies while also accounting for differences in hardware capabilities and dynamic resource consumption profiles. Further complicating matters, individual tasks within the simulation are stateful and stochastic, requiring inter-task communication and synchronization to produce accurate outcomes. We deal with these challenges through intelligent state collection and migration, active resource monitoring, and empirical evaluation of resource capabilities under changing conditions. To underscore the viability of our solution, we provide benchmarks using a production discrete event simulation that can simultaneously sustain failures, manage resource heterogeneity, and handle slowdowns while being orchestrated by our framework.

Connecting Researchers to HPCS through Web Services

Researchers may find it difficult to run experiments on High Performance Computing Systems (HPCS) if they are unfamiliar with terminals and UNIX/Linux commands. Instead of using available HPCS resources, they may choose to run experiments on their personal computers, even if it takes them weeks to run small jobs. However, these researchers are familiar with using the Internet on a daily basis to check email or read the latest news. To make high performance computing more accessible and to help researchers use the HPCS available to them, the authors have developed a web based interface for running simulations from a web browser. This paper describes the design of this interface and its benefits.

Learning Based Distributed Orchestration of Stochastic Discrete Event Simulations

Discrete event simulations (DES) are used in situations where we need to understand or describe complex phenomena. This paper describes an algorithm for dynamic orchestration of stochastic DES. To cope with long execution times in stochastic DES settings, we use MapReduce to achieve concurrent processing of the simulation on a distributed collection of machines. The proposed algorithm proactively targets imbalances between subtasks of the simulation. It achieves this by accurately predicting future execution times for map instances and apportioning processing workloads while accounting for the overheads associated with the apportioning. Our empirical benchmarks demonstrate the suitability of our scheme.