The goal of ranking and selection (R8S) procedures is to identify the best among a finite set of alternative systems evaluated by stochastic simulation, providing a probability guarantee on the quality of the solution. To solve large-scale R8S problems, especially in parallel computing platforms where variable numbers of cores might be used, it is helpful to be able to predict the simulation budget, which is almost always the dominant portion of the running time of a given procedure for a given problem. Non-trivial issues arise due to the need to estimate the system configuration. We propose a set of methods for predicting the simulation budget. Numerical results compare our predictions for several leading R8S procedures.

Discrete event simulation (DES) is traditionally used as an offline tool to help users to carry out analysis for complex systems. As real-time sensor data become more and more available, there is increasing interest of assimilating real-time data into DES to achieve on-line simulation to support real-time decision making. This article presents a data assimilation framework that works with DES models. Solutions are proposed to address unique challenges associated with data assimilation for DES. A tutorial example of discrete event road traffic simulation is developed to demonstrate the data assimilation framework as well as principles of data assimilation in general. This article makes contributions to the DES community by presenting a data assimilation framework for DES and a concrete tutorial example that helps readers to grasp the details of data assimilation for DES.

Social contact network (SCN) models the daily contacts between people in real life. It consists of agents and locations. When agents visit a location at the same time, the social interactions can be established among them. Simulations over SCN have been employed to study social dynamics such as disease spread among population. Because of the scale of SCN and the execution time requirement, the simulations are usually run in parallel. However, a challenge to the parallel simulation is that the structure of SCN is naturally skewed with a few hub locations that have far more visitors than others. These hub locations can cause load imbalance and heavy communication between partitions, which therefore impact the simulation performance. This article proposes a comprehensive solution to address this challenge. First, the hub locations are decomposed into small locations, so that SCN can be divided into partitions with better balanced workloads. Second, the agents are decomposed to exploit data locality, so that the overall communication across partitions can be greatly reduced. Third, two enhanced execution mechanisms are designed for locations and agents, respectively, to improve simulation parallelism. To evaluate the efficiency of the proposed solution, an epidemic simulation was developed and extensive experiments were conducted on two computer clusters using three SCN datasets with different scales. The results demonstrate that our approach can significantly improve the execution performance of the simulation.

We present a new approach for efficient process synchronization in parallel discrete event simulation on multicore computers. We aim specifically at simulation of spatially extended stochastic system models where time intervals between successive inter-process events are highly variable and without lower bounds: This includes models governed by the mesoscopic Reaction-Diffusion Master Equation (RDME). A central part of our approach is a mechanism for optimism control, in which each process disseminates accurate information about timestamps of its future outgoing interprocess events to its neighbours. This information gives each process a precise basis for deciding when to pause local processing to reduce the risk of expensive rollbacks caused by future “delayed” incoming events. We apply our approach to a natural parallelization of the Next Subvolume Method (NSM) for simulating systems obeying RDME. Since this natural parallelization does not expose accurate timestamps of future interprocess events, we restructure it to expose such information, resulting in a simulation algorithm called Refined Parallel NSM (Refined PNSM). We have implemented Refined PNSM in a parallel simulator for spatial extended Markovian processes. On 32 cores, it achieves an efficiency ranging between 43--95% for large models, and on average 37% for small models, compared to an efficient sequential simulation without any code for parallelization. It is shown that the gain of restructuring the naive parallelization into Refined PNSM more than outweighs its overhead. We also show that our resulting simulator is superior in performance to existing simulators on multicores for comparable models.

The multiobjective simulation optimization (MOSO) problem is a nonlinear multiobjective optimization problem in which multiple simultaneous and conflicting objective functions can only be observed with stochastic error. We provide an introduction to MOSO at the advanced tutorial level, aimed at researchers and practitioners who wish to begin working in this emerging area. Our focus is exclusively on MOSO methods that characterize the entire efficient or Pareto-optimal set as the solution to the MOSO problem; later, this set may be used as input to the broader multicriteria decision-making process. Our introduction to MOSO includes an overview of existing theory, methods, and provably convergent algorithms that explicitly control sampling error for (1) MOSO on finite sets, called multiobjective ranking and selection; (2) MOSO with integer-ordered decision variables; and (3) MOSO with continuous decision variables. In the context of integer-ordered and continuous decision variables, we focus on methods that provably converge to a local efficient set under the natural ordering. We also discuss key open questions that remain in this emerging field.

We introduce Path-ZVA: an efficient simulation technique for estimating the probability of reaching a rare goal state before a regeneration state in a (discrete-time) Markov chain. Standard Monte Carlo simulation techniques do not work well for rare events, so we use importance sampling; i.e., we change the probability measure governing the Markov chain such that transitions “towards” the goal state become more likely. To do this, we need an idea of distance to the goal state, so some level of knowledge of the Markov chain is required. In this article, we use graph analysis to obtain this knowledge. In particular, we focus on knowledge of the shortest paths (in terms of “rare” transitions) to the goal state. We show that only a subset of the (possibly huge) state space needs to be considered. This is effective when the high dependability of the system is primarily due to high component reliability, but less so when it is due to high redundancies. For several models, we compare our results to well-known importance sampling methods from the literature and demonstrate the large potential gains of our method.

Collective adaptive systems (CAS) often adopt cooperative operating strategies to run distributed decision-making mechanisms. Sometimes, their effectiveness massively relies on the collaborative nature of individuals’ behavior. Stimulating cooperation while preventing selfish and malicious behaviors is the main objective of trust and reputation models. These models are largely used in distributed, peer-to-peer environments and, therefore, represent an ideal framework for improving the robustness, as well as security, of CAS. In this article, we propose a formal framework for modeling and verifying trusted CAS. From the modeling perspective, mobility, adaptiveness, and trust-based interaction represent the main ingredients used to define a flexible and easy-to-use paradigm. Concerning analysis, formal automated techniques based on equivalence and model checking support the prediction of the CAS behavior and the verification of the underlying trust and reputation models, with the specific aim of estimating robustness with respect to the typical attacks conducted against webs of trust.

Wireless Sensor Networks (WSNs) are important examples of Collective Adaptive System, which consist of a set of motes that are spatially distributed in an indoor or outdoor space. Each mote monitors its surrounding conditions, such as humidity, intensity of light, temperature, and vibrations, but also collects complex information, such as images or small videos, and cooperates with the whole set of motes forming the WSN to allow the routing process. The traffic in the WSN consists of packets that contain the data harvested by the motes and can be classified according to the type of information that they carry. One pivotal problem in WSNs is the bandwidth allocation among the motes. The problem is known to be challenging due to the reduced computational capacity of the motes, their energy consumption constraints, and the fully decentralised network architecture. In this article, we study a novel algorithm to allocate the WSN bandwidth among the motes by taking into account the type of traffic they aim to send. Under the assumption of a mesh network and Poisson distributed harvested packets, we propose an analytical model for its performance evaluation that allows a designer to study the optimal configuration parameters. Although the Markov chain underlying the model is not reversible, we show it to be ρ-reversible under a certain renaming of states. By an extensive set of simulations, we show that the analytical model accurately approximates the performance of networks that do not satisfy the assumptions. The algorithm is studied with respect to the achieved throughput and fairness. We show that it provides a good approximation of the max-min fairness requirements.

This article focuses on evaluating the probability that both components of a two-dimensional stochastic process will ever, but not necessarily at the same time, exceed some large level u . An important application is in determining the probability of large delays occurring in two correlated queues. Since exact analysis of this probability seems prohibitive, we focus on deriving asymptotics and on developing efficient simulations techniques. Large deviations theory is used to characterise logarithmic asymptotics. The second part of this article focuses on efficient simulation techniques. Using “nearest-neighbour random walk” as an example, we first show that a “naive” implementation of importance sampling, based on the decay rate, is not asymptotically efficient. A different approach, which we call partitioned importance sampling, is developed and shown to be asymptotically efficient. The results are illustrated through various simulation experiments.

Simulation is commonly used to study the random behaviors of large-scale stochastic systems with correlated inputs. Since the input correlation is often induced by latent common factors in many situations, to facilitate system diagnostics and risk management, we introduce a factor-based Bayesian framework that can improve both computational and statistical efficiency and provide insights for system risk analysis. Specifically, we develop a flexible Gaussian copula-based multivariate input model that can capture important properties in the real-world data. A nonparametric Bayesian approach is used to model marginal distributions, and it can capture the properties, including multi-modality and skewness. We explore the factor structure of the underlying generative processes for the dependence. Both input and simulation estimation uncertainty are characterized by the posterior distributions. In addition, we interpret the latent factors and estimate their effects on the system performance, which could be used to support diagnostics and decision making for large-scale stochastic systems. Our approach is supported by both asymptotic theory and empirical study.