Keywords: Grid computing technology, Grid middleware, 3D structural dynamic analysis, large dimension buildings, HPC techniques, direct time integration methods.

This paper describes a Grid Computing [1] application for the 3D dynamic analysis of large buildings that aims at using a distributed computational infrastructure in order to accelerate the simulation of case studies which are composed of different structural solutions where one or more dynamic loads are applied to them.

A previously developed parallel simulator has been integrated into this system. It carries out a 3D linear dynamic analysis of buildings and it is composed of parallel MPI-based implementations of eight well-known direct time integration methods [2]. Static condensation of nodes has not been assumed and all the elements and nodes of the structure are taken into account during the simulation process. Also, six degrees of freedom per each node are considered.

To analyse the response of a building under the influence of different earthquakes, multiple simulations must be executed which, depending on the magnitude of the building, may be time-consuming as well as memory-intensive for a traditional PC. Thus, performing multiple concurrent executions on the computational resources of a Grid deployment could be an ideal solution to accelerate the execution of whole resource-starved structural studies.

In order to harness the power that a Grid infrastructure aims to deliver, we have developed a software application that enables structural dynamic simulations to be perfomed on the computational nodes of a Grid infrastructure. For the development of such an application we have employed GMarte [3], a software layer developed by our research group that enables the simplification the process of remote task execution of scientific applications on a Globus-based Grid. GMarte is a user-level middleware which provides an object-oriented, high level view of the Grid. The underlying Grid middleware employed is the industrial standard Globus Toolkit 2.4.3 [4].

The Grid-based application provides meta-scheduling functionality thus allowing performing automatic task allocation to the resources of a Grid deployment. This procedure involves the following set of steps: For each structural alternative to be simulated, the most appropriate resource is selected. Then, all the required input files are transferred to the remote machine. Once this phase has been completed, the parallel structural simulator is remotely launched. When the execution finishes, all the generated output files are moved to the local machine, erasing all the remote archives. The application is provided with fault-tolerance capabilities, so if the execution fails, the task will be re-scheduled.

To test the effectiveness of the Grid application, a structural case study has been executed on a Grid deployment composed of three clusters of PCs from our research group. The case study considers an apartment building composed of two basements, one ground floor and thirty-three stories. The building is composed of reinforced concrete, and eight combinations of different structural member dimensions were considered for its design. Each structural alternative was composed of about 40,000 structural elements and 22,000 nodes, i.e. 132,000 degrees of freedom. Taking into account the current Spanish-Resistant Construction Standards, each of the eight different preliminary solutions was linearly simulated under the influence of five different representative earthquakes according to the geographical location of the building. This case study results in 40 independent structural dynamic simulations.

On the one hand, the Grid-based simulation process of all the structural solutions required 34 minutes from the beginning of the scheduling until the data transfer for the last task finished. On the other hand, a traditional sequential execution of the whole case study on a single Pentium Xeon at 2.8 GHz. with 1 GByte of RAM lasted a total of 354 minutes (almost 6 hours). A High Performance Computing approach performing two concurrent four-processor executions took 63.73 minutes. The Grid computing approach was more than 10 times faster than using a traditional sequential execution and almost 2 times faster than using the parallel computing alternative. Generalized- direct integration method was employed in all the mentioned dynamic analyses.

1

I. Foster, C. Kesselman, "The GRID: Blueprint for a New Computing Infrastructure", Morgan-Kaufmann, 1999.

2

T.C. Fung, "Numerical Dissipation in Time-Step Integration Algorithms for Structural Dynamic Analysis", Progress in Structural Engineering and Materials, 5, 167-180, 2003. doi:10.1002/pse.149

3

J.M. Alonso, V. Hernández, G. Moltó, "An Object-Oriented View of Grid Computing Technologies to Abstract Remote Task Execution", 13th Euromicro Conference on Parallel, Distributed and Network-based Processing (PDP 2005), 235-242, IEEE, 2005. doi:10.1109/EMPDP.2005.12

4

I. Foster, C. Kesselman, "Globus: A MetaComputing Infrastructure Toolkit", International Journal of Supercomputer Applications, 11, 115-128, 1997. doi:10.1177/109434209701100205

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £135 +P&P)