Average Reviews:
(More customer reviews)The computational Aspects of General Relativity, by J. Frauendiener, is one of many rather interesting papers one would find in this book. This book covers the 2nd Russian-German Advanced Research Workshop which took place in March of 2005. The papers that were presented cover the applications of High Performance Computing (HPC) in research areas such as Fluid Mechanics and computational Physics. As one would expect, the underlying pattern of the types of problems faced by these area are the same as any other, mainly they must be parallelizable, but the effort taken by various researchers to essentially breakup a given problem is truly remarkable.
Particle methods in powder technology, by B. Henrich, et. al., is another paper which the authors use the HPC methods to simulate compaction, sintering and filling of a dispersed powder. The simulation, as one would expect, uses a molecular dynamics method which in turn follows Newton's equations of motions for simulating the time evolution of particles. As one would expect, the sheer number of simulations that must take place is simply staggering, and an HPC is required to get the job done is a reasonable amount of time.
There are a number of other paper that talk about the applications of HPC in Fluid Dynamics or Mechanics, but I was very interested in H. Brunst, et. al. paper, Parallel applications on large scale systems: getting insights, which the authors talk about the "scalability properties on moderns parallel computer architectures". The authors evaluate BenchIT, which is used for evaluation and presentation of performance data. BenchIT is used to compare difference architectures and programming languages against each other to determine their efficiency and performance numbers. BenchIT can be used for simply estimating what the gain would be if for example one programming language is used over the other given certain scenarios. It is a great tool for ROI (return on investments) calculations.
U. Kuster, et. al. paper, Sustaining performance in future vector processors, is simply fascinating! The authors present current performance numbers from a number of platforms, and tie the performance to the software architecture - which is very much true. Teraflops/sec performance is desired, but the authors argue that with current processor architecture, the only way to achieve such performance numbers one much re-architect the way we tackle the problem from the software architecture perspective. The authors then close the loop with possible improvements to the processor architecture that can aid in making the software design a bit simpler. The goal of this paper is show that software and hardware are more tied together than ever, and in order to "go to the next level", both need to work together instead of dispersed components.
I was very surprised at the quality of the papers that are in this text. As an architect that deals with Grid and HPC systems on daily basis, I found this text very informative and educational.
Click Here to see more reviews about: Computational Science and High Performance Computing II: The 2nd Russian-German Advanced Research Workshop, Stuttgart, Germany, March 14 to 16, 2005 ... and Multidisciplinary Design) (v. 2)
0 comments:
Post a Comment