662 research outputs found
Sort by
No full textPoznańMany large-scale parallel scientific and engineering applications, especially climate modeling, often run for lengthy periods and require data checkpointing periodically to save the state of the computation for a program restart. In addition, such applications need to write data to disks for post-processing, e.g., visualization. Both these scenarios involve a write-only pattern using Hierarchal Data Format (HDF) files. In this paper, we study the scalability of CXFS by HDF based Structured Adaptive Mesh Refinement (AMR) application for three different block sizes. The code used is a block-structured AMR hydrodynamics code that solves compressible, reactive hydrodynamic equations and characterizes physics and mathematical algorithms used in studying nuclear flashes on neutron stars and white dwarfs. The computational domain is divided into blocks distributed across the processors. Typically, a block contains 8 zones in each coordinate direction (x, y, and z) and a perimeter of guard cells (in this case, 4 zones deep) to hold information from the neighbours. We used three different block sizes of 8 × 8 × 8, 16 × 16 × 16, and 32 × 32 × 32. Results of parallel I/O bandwidths (checkpoint file and two plot files) are presented for all three-block sizes on a wide range of processor counts, ranging from 1 to 508 processors of the Columbia system
No full textPoznańIn the 3rd quarter 2005 the Technische Universität Dresden started to install the first phase of a new SGI supercomputer and a PC farm build by LinuxNetworX. The final phase of the installation will take place in July/August 2006. One main focus of the installation is to have not just outstanding I/O-Performance for the disk to memory channel but for the tape to disk channel too. In this paper we will show how our setup looks like and what our expectations for the performance are. We will also present some measurements done on the system and show the performance boundaries of the current setup
No full textPoznańThe Columbia system at the NASA Advanced Supercomputing (NAS) facility is a cluster of 20 SGI Altix nodes, each with 512 Itanium 2 processors and 1 terabyte (TB) of shared-access memory. Four of the nodes are organized as a 2048-processor capability computing platform connected by two low-latency interconnects – NUMALink4 (NL4) and InfiniBand (IB). To evaluate the scalability of Columbia with respect to both increased processor counts and increased problem sizes, we used seven of the NAS Parallel Benchmarks and all three of the NAS multi-zone benchmarks. For NPB we ran three Classes B, C, and D of benchmarks. To measure the impact of some architectural features, we compared Columbia results with results obtained on a Cray Opteron Cluster consisting of 64 nodes, each with 2 AMD Opteron processors and 2 gigabytes (GB) of memory, connected with Myrinet 2000. In these experiments, we measured performance degradation due to contention for the memory buses on the SGI Altix BX2 nodes. We also observed the effectiveness of SGI’s NL4 interconnect over Myrinet. Finally, we saw that computations spanning multiple BX2 nodes connected with NL4 performed well. Some computations did almost as well when the IB interconnects was used
No full textPoznańA conducting half-space, permeated by an initial magnetic field governed by the generalized equations of thermoelasticity is considered. The bounding plane is acted upon by a combination of thermal and mechanical shock. The formulation is applied to both generalizations, Lord-Shulman theory and the Green-Lindsay theory, as well as to the coupled theory. Laplace transform techniques together with the method of potentials are used. The inversion of the Laplace is carried out using a numerical approach. Numerical results for the temperature, the stress and the induced magnetic and electric field distributions are obtained and illustrated graphically for a particular case. Comparisons are made with the results obtained in the case of the absence of the magnetic field
No full textPoznańA modified algorithm of the numerical calculation of adiabatic corrections is proposed. It is based on earlier approaches, introduced by Cencek and Kutzelnigg and by the present author. The adiabatic correction is approximated in terms of overlap integrals between electronic wave functions for a given geometry and for a single nucleus shifted by h. The leading term of error is proportional to the square of h in the original methods. The new approach, while requires the same wave functions as those methods, shrinks the error, so that it becomes proportional to h4 in atomic cases. Test calculations show, that similar behavior is retained also for two–atom molecules and additional stable decimal digit of the adiabatic correction can be obtained
No full textPoznańIn the paper we present two different models of Grid resource management problems: (i) Grid scheduling problems with no time characteristics available, and (ii) scheduling of jobs in presence of time characteristics achieved by using some prediction techniques, and resource reservation mechanisms. We focus on demonstrating how these two scenarios, which are important examples of Grid environments, can be modeled as multi-criteria decision support problems. We also discuss advantages and disadvantages of these models as well as practical applications
