DSMC: Fast direct simulation Monte Carlo solver for the Boltzmann equation by Multi-Chain Markov Chain and multicore programming

Direct Simulation Monte Carlo (DSMC) solves the Boltzmann equation with large Knudsen number. The Boltzmann equation generally consists of three terms: the force term, the diffusion term and the collision term. While the first two terms of the Boltzmann equation can be discretized by numerical methods such as the finite volume method, the third term can be approximated by DSMC, and DSMC simulates the physical behaviors of gas molecules. However, because of the low sampling efficiency of Monte Carlo Simulation in DSMC, this part usually occupies large portion of computational costs to solve the Boltzmann equation. In this paper, by Markov Chain Monte Carlo (MCMC) and multicore programming, we develop Direct Simulation Multi-Chain Markov Chain Monte Carlo (DSMC3): a fast solver to calculate the numerical solution for the Boltzmann equation. Computational results show that DSMC3 is significantly faster than the conventional method DSMC.

A Study of Parallelization and Performance Optimizations Based on OpenMP

The primary consequence of the transition to multicore processors is that applications will increasingly need to be parallelized to improve application's throughput, responsiveness and latency. Multithreading is becoming increasingly important for modern programming. Unfortunately, parallel programming is no doubt much more tedious and error-prone than serial programming. Although modern compilers can manage threads well, but in practice, synchronization errors (such as: data race errors, deadlocks) required careful management and good optimization method. This paper presents a preliminary study of the usability of the Intel threading tools for multicore programming. This work compare performance of a single threaded application with multithreaded applications, use tools called Intel® VTune Performance Analyzer, Intel® Thread Checker and OpenMP to efficiently optimize multithreaded applications.

High-level multicore programming with C++11

Nowadays, one of the most important challenges in programming is the efficient usage of multicore processors. All modern programming languages support multicore programming at native or library level. C++11, the next standard of the C++ programming language, also supports multithreading at a low level. In this paper we argue for some extensions of the C++ Standard Template Library based on the features of C++11. These extensions enhance the standard library to be more powerful in the multicore realm. Our approach is based on functors and lambda expressions, which are major extensions in the language. We contribute three case studies: how to efficiently compose functors in pipelines, how to evaluate boolean operators in parallel, and how to efficiently accumulate over associative functors.