Keywords: earthquake engineering, numerical simulation, software integration, high performance computing, software engineering, object oriented design.

An Integrated Earthquake Simulation (IES) environment as a candidate for numerical computation based hazard and disaster prediction environments was introduced by authors in previous studies [1,2,3]. In order to enhance IES with HPC, two improvements are discussed in this paper: 1) layer-based design of the IES architecture and Common Modeling Data (CMD) for inter-layer communication; and 2) Generic Distributable Class (GDC) and an MPI_Process_Manager class for sending objects with a complicated structure and massive data through the MPI.

The layer-based design is described in terms of one kernel and four layers, namely, data, hazard simulation, disaster simulation and visualization layers; each layer consists of aggregated abstract classes.

For IES the use of the standard Message Passing Interface library is not feasible, because the CMD has complicated data structure and the data stored in the CMD cannot be transferred by using the MPI library as well as its extensions [4]. Therefore to cope with this issue, the authors are developing GDC. GDC is an abstract class which offers two functionalities, serialization and de-serialization which are overloading MPI functions of Pack and Unpack with specific consideration made with respect to the complex data structure of the CMD.

The MPI_Process_Manager class is introduced to make easy use of the MPI library in IES, achieving the implementation of a parallel computation environment to the IES in a general manner so that this class could be used for other simulation systems. A client-server model is selected for the parallel computing architectural model, and first-come-first-served (FCFS) is used as a load balancing strategy. Two analysis methods for linear and nonlinear seismic response analysis are used to measure the performance of HPC enhanced IES, namely, linear multi-degree-of-freedom (MDOF) analysis and non-linear distinct element method (DEM) analysis. The performance is measured in terms of speedup and efficiency, and scalability is estimated using iso-efficiency [5] functions and index.

The speedup measure shows a good sign of performance when the number of structures increases and respectively when a more advance simulation program is used, such as the DEM. The estimated iso-efficiency function is of low power, and hence it is shown that the present IES has scalability from degree two. The iso-efficiency functions show that IES scalability is independent from the analysis methods, even though the efficiency of the function is 0.25 and 0.47 for the MDOF and DEM analyses, respectively. It can be expected that IES has similar good efficiency for other analysis methods.

1

M. Hori, "Introduction to Computational Earthquake Engineering," Imperial College Press, 2006.

2

S. Nallathamby, M. Hori, Y. Ariga, "Improvement of IES Wrapper for Plug-in of Advanced Simulation Programs", Journal of Applied Mechanics, 10, 573-579, 2007.

3

T. Ichimura, M. Hori, "Development of prototype of Integrated Earthquake Disaster Simulator using digital city and strong ground motion simulator with high resolution," 13th Conference on Earthquake Engineering, No. 1418, 2004.

4

B.C. McCandless, J.M. Squyres, A. Lumsdaine, "Object Oriented MPI (OOMPI), a class library for the Message Passing Interface," MPI Developer's Conference, Proceedings, 87-94, 1996.

5

P.S. Pacheco, "Parallel programming with MPI," Morgan Kaufmann Publishers Inc., 1996.

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 £72 +P&P)