**5.2.1 Porting and optimization of codes**

Preliminary studies have been devoted to expose seismologists and seismic engineers to modern e-infrastructures (which includes both HPC and Grid environments) so that the potential provided by this infrastructure for seismic hazard assessment research can be assessed and exploited. These activities aim to enable the computational seismology user community to the use of modern e-infrastructure and acquire the core innovations emerging in this framework, for example the development of an European and worldwide einfrastructure for advanced applications in seismic hazard assessment driven by European Union projects such as EU-IndiaGRID2. The major goals of this new collaboration are to:


Activities carried out so far have been dedicated to a general introduction to the e-Infrastructures for Grid and HPC and to the preliminary assessment of their use in seismological research, with special emphasis on methods for advanced definition of ground shaking scenarios based on physical modelling of seismic waves generation and propagation processes. Researchers gained some practice on the use of the e-infrastructure for neo-deterministic seismic hazard assessment at different scales and level of detail, working actively on Italian data and testing the specialized seismological software running

Applications Exploiting e-Infrastructures Across

0

3.75

7.5

15

11.25

Hours

Europe and India Within the EU-IndiaGrid Project 303

Fig. 6. Speedup of seismograms computation obtained on our local machine

the results and relaunching aborted jobs were developed.

**5.2.2 Preliminary results** 

ground shaking peaks at each receiver.

The second step was porting the optimized hazard package on EU-IndiaGrid infrastructure. Two different types of parametric tests were developed: on the deterministic source parameters and on the random properties of the source model. The first experiment is performed by perturbing the properties of the seismic sources selected by the algorithm before the computation of synthetic seismograms. In the second test different sets of curves of source spectrum generated by a MonteCarlo simulation of the source model are used for scaling the seismograms. In both cases there are many independent runs to be executed, so a script for generation of the input and other scripts for checking the status of jobs, retrieving

Before After I/O opt After sorting

Two preliminary tests over deterministic source parameter for a restricted area ("persut") and for whole Italy ("persut Italy"), and two different tests over random properties ("seed1Hz" and "seed10Hz") for the whole Italian territory, with different frequency content and different maximum distance for the computation of seismograms, were conducted. So the performance of the package over the grid in terms of computational time and number of successful jobs were tested, and submission of job and retrieval of its output were refined. The number of seismograms that must be computed determines the duration and the storage requirement of the run. This parameter seems critical for the success of the job. The test runs on the random component of the source gave an indication on the effective number of jobs that must be computed to have a good estimate of the distribution of the

on an e-Infrastructure environment, leveraging on the work performed within the EU-IndiaGrid projects.

The use of the EU-India Grid infrastructure allows conducting massive parametric tests, to explore the influence not only of deterministic source parameters and structural models but also of random properties of the same source model, to enable realistic estimate of seismic hazard and their uncertainty. The random properties of the source are especially important in the simulation of the high frequency part of the seismic ground motion.

We have ported and tested seismological codes for national scale on the Grid infrastructure Flowchart of this code is illustrated in figure 5. The first step of the work, performed at a EU-IndiaGrid side-training event was the optimization of whole package by the identification of the critical programs and of the hot spots within programs. The critical point in the algorithm was the computation of synthetic seismograms. The optimization was performed in two ways: first by removing of repeated formatted disk I/O, second by sorting of seismograms by source depth, to avoid the repeated computation of quantities that are depth-dependent. Figure 6 show the remarkable improvement obtained in such work.

Fig. 5. Flow chart of national scale hazard package

302 Grid Computing – Technology and Applications, Widespread Coverage and New Horizons

on an e-Infrastructure environment, leveraging on the work performed within the EU-

The use of the EU-India Grid infrastructure allows conducting massive parametric tests, to explore the influence not only of deterministic source parameters and structural models but also of random properties of the same source model, to enable realistic estimate of seismic hazard and their uncertainty. The random properties of the source are especially important

We have ported and tested seismological codes for national scale on the Grid infrastructure Flowchart of this code is illustrated in figure 5. The first step of the work, performed at a EU-IndiaGrid side-training event was the optimization of whole package by the identification of the critical programs and of the hot spots within programs. The critical point in the algorithm was the computation of synthetic seismograms. The optimization was performed in two ways: first by removing of repeated formatted disk I/O, second by sorting of seismograms by source depth, to avoid the repeated computation of quantities that are depth-dependent. Figure 6 show the remarkable improvement obtained in such work.

in the simulation of the high frequency part of the seismic ground motion.

Fig. 5. Flow chart of national scale hazard package

IndiaGrid projects.

Fig. 6. Speedup of seismograms computation obtained on our local machine

The second step was porting the optimized hazard package on EU-IndiaGrid infrastructure. Two different types of parametric tests were developed: on the deterministic source parameters and on the random properties of the source model. The first experiment is performed by perturbing the properties of the seismic sources selected by the algorithm before the computation of synthetic seismograms. In the second test different sets of curves of source spectrum generated by a MonteCarlo simulation of the source model are used for scaling the seismograms. In both cases there are many independent runs to be executed, so a script for generation of the input and other scripts for checking the status of jobs, retrieving the results and relaunching aborted jobs were developed.
