China
Africa
Explore chapters and articles related to this topic
Parallel Computing Models
Published in Vivek Kale, Parallel Computing Architectures and APIs, 2019
The random-access machine (RAM) model captures the essential features of traditional sequential computers. The RAM model is composed of a single processor and a memory with sufficient capacity; each memory location can be accessed in a random (direct) way. In each time step, the processor performs one instruction as specified by a sequential algorithm. Instructions for (read or write) access to the memory as well as for arithmetic or logical operations are provided. The RAM model provides a simple model that abstracts from many details of real computers, such as a fixed memory size and the existence of a memory hierarchy with caches, complex addressing modes, or multiple functional units; however, the RAM model can also be used to perform a runtime analysis of sequential algorithms to describe their asymptotic behavior, which is meaningful for real sequential computers.
Parallel algorithm development and testing using Petri-object simulation
Published in International Journal of Parallel, Emergent and Distributed Systems, 2021
Inna V. Stetsenko, Alexander A. Pavlov, Oleksandra Dyfuchyna
The development of a parallel algorithm is a hard task that often requires fundamental reconstruction of the sequential algorithm. The programmer tries to exploit the computing resources in the most effective way to achieve the highest speed-up. However, the usage of synchronised processes and the cost of multithreading support can ruin his efforts. Synchronisation will always slow down the performance of the algorithm. Threads will always need additional resource-consuming operations to control its running. Thus, the programmer should find a balance between design complexity, threads’ resource consumption and context switching overhead, and be sure that the multithreading costs are less than multithreading benefits [14].
An approach to extracting surface supply relationships between glaciers and lakes on the Tibetan Plateau
Published in International Journal of Digital Earth, 2018
Bei-Bei Ai, Cheng-Zhi Qin, Qinghua Ye, A-Xing Zhu, Graham Cogley
The acceleration performance of the parallel algorithm for the first stage in the proposed approach was evaluated based on runtime, speedup ratio, and parallel efficiency. The runtime included only the execution time for building the direct connections by digital terrain analysis, excluding the time needed for I/O between internal and external memory. It would take too long to build the direct connections for the whole Tibetan Plateau using either the sequential algorithm or the parallel algorithm with a single process. Therefore, the acceleration performance of the parallel algorithm was tested in one representative basin.
Convergence Studies on Nonlinear Coarse-Mesh Finite Difference Accelerations for Neutron Transport Analysis
Published in Nuclear Science and Engineering, 2018
In this paper, the convergence rates of the two CMFDs were evaluated for the sequential algorithm and the parallel algorithm. For the neutron transport calculation, the discrete ordinates method with the S16 Gauss-Legendre quadrature set was used. The tests were performed in both analytic and numerical approaches, and the well-known Fourier analysis was applied for analytic convergence study. Since this paper focused on the performance of the CMFDs in a parallel computing environment, the Fourier analysis of the parallel CMFD acceleration was described in detail.