A fully distributed CFD framework for massively parallel systems
نویسندگان
چکیده
A highly scalable solver framework, based on a linearized octree is presented. It allows for fully distributed computations and avoids special needs on single processes with potential bottlenecks, while enabling simulations with complex geometries. Scaling results on the Cray XE6 system Hermit at HLRS in Stuttgart are presented with runs up to 3072 nodes with 98304 MPI processes. Even with a fully indirect addressing a high sustained performance of more than 9% can be achieved on the system, enabling very large simulations. Two flow simulation methods are shown, a Finite Volume Method for compressible flows, and a Lattice Boltzmann Method for incompressible flows in complex geometries. Keywords-Octree; Computational fluid dynamics; Finite Volume; Lattice Boltzmann; High Performance Computing; Performance; Scaling
منابع مشابه
Static Task Allocation in Distributed Systems Using Parallel Genetic Algorithm
Over the past two decades, PC speeds have increased from a few instructions per second to several million instructions per second. The tremendous speed of today's networks as well as the increasing need for high-performance systems has made researchers interested in parallel and distributed computing. The rapid growth of distributed systems has led to a variety of problems. Task allocation is a...
متن کاملThe Legacy and Future of Cfd at Los Alamos
The early history is presented of the prolific development of CFD methods in the Fluid Dynamics Group (T-3) at Los Alamos National Laboratory in the years from 1958 to the late 1960's. Many of the currently used numerical methods –PIC, MAC, vorticity-stream-function, ICE, ALE methods and the k-ε method for turbulence– originated during this time. The rest of the paper summarizes the current res...
متن کاملJOVE: A Dynamic Load Balancing Framework for Adaptive Computations on an SP-2 Distributed-Memory Multiprocessor
Computational requirements of full scale computational fluid dynamics change as computation progresses on a parallel machine. The change in computational intensity causes workload imbalance of processors, which in turn requires a large amount of data movement at runtime. If parallel CFD is to be successful on a parallel or massively parallel machine, balancing of the runtime load is indispensab...
متن کاملAiaa - 92 - 2734 - Cp Future Directions in Computing and Cfd
generations of computers as well as improvements in algorithms. As a results CFD has become an ever more powerful tool in the design of aerospace systems. The high computational demands of aeroscience applications have been the driving force behind the recent rapid development of CFD. Moreover, CFD has been an equally successful modeling tool in a variety of other fields, such as automotive eng...
متن کاملSoftware Engineering Methods for Parallel and Distributed Scientific Computing
In this paper, we present an interdisciplinary research project whose central objective is to develop new software engineering (SWE) methods for distributed memory parallel scientific computing. Our emphasis is on putting into practice and evaluating the proposed methods. The main test case for their definition and evaluation is the parallelization of an industrial CFD software package. A major...
متن کامل