Preface
This work is an effort to produce a text that is helpful for those who are becoming engineers, and for postgraduate students and information technology (IT) professionals. Without doubt, Computational Fluid Dynamics (CFD) is an emerging technology advancing with the arrival of modern supercomputers. Therefore, knowledge of CFD alone is not sufficient to compete with ongoing challenges in this field. Many engineers are handicapped when dealing with IT matters and they wait a long time for an IT expert to solve problems. Therefore, to a certain extent, knowledge of High-Performance Computing (HPC) is mandatory for engineers. From a different perspective, when IT professionals are dealing with CFD problems, they need to know the basics of most ongoing commercial codes, especially their installation on Linux-based clusters, which enables efficient job scheduling and troubleshooting bottlenecks in running CFD software without the help of CFD engineers. It is important that the reader not feel the need to browse the Internet for simple matters which this book attempts to explain more easily. This book thus focuses on two aspects: HPC and CFD.
In the initial chapters, we concentrate on CFD aspects, its rationale, and advanced numerical schemes. Later chapters focus on HPC and the way HPC works in CFD, and then switch to different software variations as they are used in HPC, with a particular focus on commercial code ANSYS Fluent benchmarks.
Chapter 1 introduces CFD. Many organizations implement CFD in the computer-aided engineering phase. However, most of the time, higher management is not interested, perhaps because of lengthy simulations or the uncertainty of results. These issues are discussed and various misconceptions about CFD are explored and clarified. The basics of CFD with governing equations are also discussed.
Chapter 2 introduces HPC. The pioneers of HPC and their contributions are discussed and the world Top 5 computers are mentioned and discussed in detail.
Chapter 3 focuses on CFD algorithms designed for parallel machines. This is basically the treatment of CFD codes for parallel processing, because CFD codes designed for traditional serial machines may not run efficiently on parallel computers. This chapter focuses on how CFD can be used in parallel and on parallel architectures, with some discussion of mesh partitioning techniques.
Chapter 4 discusses turbulence and its high Reynolds representations, such as Direct Numerical Simulations (DNS), Large Eddy Simulations (LES), and Detached Eddy Simulations. The methods are not discussed from a mathematical point of view, to maintain the reader’s interest. Some examples of DNS and LES are given and discussed: particularly, free shear flows and wall-bounded flows. Computational requirements in floating point operations per second with respect to the Reynolds number are mentioned, which is certainly useful if the user wants an idea of cost estimation for his or her machine, keeping in mind the size of the problem. Detached eddy simulation rationales are touched on and not discussed in detail because the method is not sensitive to very high computational power compared with LES and DNS. The chapter concludes with a discussion on the importance of each method.
Chapter 5 contains mainly a classification of clusters. Many types of clusters are used globally. It all depends on the needs of the user and the budget the solution requires. This chapter gives an idea about different types of distributed clusters and their advantages. The qualities of clusters and the different components of clusters necessary to construct an HPC system are also discussed in detail. The details of chassis-based servers and rack-mounted and desktop-based clusters such as CRAY CX1 are also discussed.
Chapter 6 is fairly large chapter that deals mainly with the onboard usage of HPC in commercial CFD codes. This is a detailed chapter that explains the memory requirement for ANSYS Fluent based on the problem size, and then interconnectivity, which has a vital role in HPC, is discussed with a focus on Infiniband. Storage requirements are also highlighted. After that, Fluent benchmarks are discussed with benchmarking of medium- to high-level problems. Other codes such as ANSYS CFX and OpenFOAM are also described from a benchmark point of view. A flowchart is provided so that the user can easily purchase or form an HPC machine. A comparison of different types of machines is included to give the user an idea as to how a machine would be suitable for his or her needs.
The performance of world-famous supercomputer t-Platforms is discussed. It is a leading HPC company in Russia. With headquarters in Germany, the company is credited for establishing one of the biggest Russian clusters, named Lomonosov, at Moscow State University. I mention and discuss in detail the Lomonosov cluster and its setup, high-power availability, and performance. In addition to Lomonosov, small clusters of t-Platforms are mentioned and described briefly.
Chapter 7 discusses HPC from a networking point of view. Transmission Control Protocol/Internet Protocol, Internet Protocol addressing, ssh, and remote access via PuTTy and WinSCP for running Fluent jobs are discussed in detail. Configuring Fluent on an Windows HPC server is described.
The last chapter discusses Graphics Processing Units (GPUs), an emerging technology in the HPC and graphics fields. Apart from graphics support, GPUs are able to perform routine calculations like an ordinary CPU, but several times faster and in a more efficient and optimized way. The CUDA architecture, which acts like the life blood of GPUs, is discussed in detail in this chapter. Two generations of NVIDIA GPUs, NVIDIA Tesla and GT200, are covered. Finally, with a focus on CFD applications, the use of GPU in CFD is discussed, mentioning the research of different scientists and engineers. This effort has been done to facilitate the work of engineers, scientists, and IT professionals, to serve as a meaningful and interesting doctrine in the field of CFD and HPC.
I hope that this text will be beneficial for the CFD engineers, scientists, and HPC professionals who are keen to learn about HPC and CFD and link the two. High-performance computing is the present and the future; one must be familiar with all of its basic needs and prerequisites. It surely has room for improvement, and the reader’s response is highly welcome, but it is also hoped that this text will quench the thirst of those who want to excel in both research areas.
Karachi, December 2014
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.