Leonardo Ramirez Guzman | Universidad Nacional Autónoma de México (original) (raw)

Papers by Leonardo Ramirez Guzman

The Wasatch Fault bounds the Salt Lake Basin (SLB), Utah to the east and is capable of producing ... more The Wasatch Fault bounds the Salt Lake Basin (SLB), Utah to the east and is capable of producing M7 earthquakes. In order to characterize the seismic hazard that such earthquakes pose to Salt Lake City and other heavily populated regions along the Wasatch Front, we require estimates of potential earthquake ground motions. However, the long recurrence time of these events (~1.3ky) necessitates that these estimates be derived from empirical relations observed in other regions or from deterministic modeling. Although previous and on-going studies (Liu et al., 2010; Roten et al., 2010) have begun to examine the earthquake ground motions caused by simulated M7 events on the Salt Lake Segment of the Wasatch Fault, a comprehensive validation of the Wasatch Community Velocity Model (WCVM) (Magistrale et al., 2006) has not yet been performed. We present the results from systematic testing of the intrinsic attenuation and seismic wave speed relations in the WCVM. Modifications to the empirical wavespeed relations and selection of intrinsic attenuation relations for the WCVM version 3c are made to minimize the differences between simulated and observed seismograms from small (M<5) earthquakes in the Wasatch Front. Simulations are carried out with the Hercules finite element tool-chain (Tu et al., 2006) and are valid at frequencies up to 1 Hz. We find a good agreement between body wave arrival times from the observed and simulated events. Discrepancies in the peak ground velocities and waveform durations are reduced in the final model. Some unresolved differences in the observed and synthetic seismograms persist and may result from un-modeled heterogeneities outside of the Salt Lake Basin. Future modeling efforts in the Wasatch Front may apply these results for improved earthquake ground motion estimates.

Abstract Conventional parallel scientific computing uses files as interface between simulation co... more Abstract Conventional parallel scientific computing uses files as interface between simulation components such as meshing, partitioning, solving and visualizing. This approach results in time-consuming file transfers, disk I/O and data format conversions ...

I present two numerical techniques to solve the Dynamic problem. I revisit and modify the Split N... more I present two numerical techniques to solve the Dynamic problem. I revisit and modify the Split Node approach and introduce a Stress Glut type Method. Both algorithms are implemented using a iso/sub- parametric FEM solver. In the first case, I discuss the formulation and perform an analysis of convergence for different orders of approximation for the acoustic case. I describe the algorithm of the second methodology as well as the assumptions made. The key to the new technique is to have an accurate representation of the traction. Thus, I devote part of the discussion to analyze the tractions for a simple example. The sensitivity of the method is tested by comparing against Split Node solutions.

Parallel supercomputing has traditionally focused on the inner kernel of scientific simulations: ... more Parallel supercomputing has traditionally focused on the inner kernel of scientific simulations: the solver. The front and back ends of the simulation pipeline---problem description and interpretation of the output---have taken a back seat to the solver when it comes to attention paid to scalability and performance, and are often relegated to offline, sequential computation. As the largest simulations move beyond

We have developed a novel analytic capability for scientists and engineers to obtain insight from... more We have developed a novel analytic capability for scientists and engineers to obtain insight from ongoing large-scale parallel unstructured mesh simulations running on thousands of processors. The breakthrough is made possible by a new approach that visualizes partial differential equation (PDE) solution data simultaneously while a parallel PDE solver executes. The solution field is pipelined directly to volume rendering, which is computed in parallel using the same processors that solve the PDE equations. Because our approach avoids the bottlenecks associated with transferring and storing large volumes of output data, it offers a promising approach to overcoming the challenges of visualization of petascale simulations. The submitted video demonstrates real-time on-the-fly monitoring, interpreting, and steering from a remote laptop computer of a 1024processor simulation of the 1994 Northridge earthquake in Southern California.

We demonstrate a new scalable approach to real-time monitoring, visualization, and steering of ma... more We demonstrate a new scalable approach to real-time monitoring, visualization, and steering of massively parallel simulations from a personal computer. The basis is an endto-end approach to parallel supercomputing in which all components -meshing, partitioning, solver, and visualization -are tightly coupled and execute in parallel on a supercomputer. This approach avoids bottlenecks associated with transfer and storage of massive simulation outputs, thereby enabling real-time visualization and steering on supercomputers with thousands of processors. We have incorporated this methodology into a framework named Hercules, which targets octree-based finite element simulations. The submitted video demonstrates real-time monitoring and steering from a laptop PC of a 1024-processor simulation of the 1994 Northridge earthquake in Southern California. Because this end-to-end approach does not require moving large data over the network and is completely scalable, our approach shows promise for overcoming the challenges of visualization of petascale simulations.

ABSTRACT Three-dimensional simulations of earthquake ground motion in sedimentary valleys have le... more ABSTRACT Three-dimensional simulations of earthquake ground motion in sedimentary valleys have led to a deeper understanding of wave propagation and site effects in urban regions. In this paper we present a preliminary study of the ground motion and resulting ...

Geophysical Journal International, 2010

We present a hybrid approach for solving the dynamic rupture problem in an anelastic medium. It c... more We present a hybrid approach for solving the dynamic rupture problem in an anelastic medium. It combines the advantages of the boundary integral equation method (BIEM), which is capable of representing accurately the solution near a crack tip, with those of the domain finite element method (FEM), which can handle conveniently heterogeneous materials, as well as the traction-free condition on a free surface and the continuity of traction across interfaces. When applied jointly, the proposed BIEM and the FEM can be used to solve efficiently spontaneous rupture problems in heterogeneous media, provided the rupture surface is contained within a homogeneous portion of the domain. The proposed method, BIEM-FEM approach (BDM, for Boundary/Domain Method), is verified for several antiplane (SH) and in-plane (P-SV) 2-D cases, in which the slip is prescribed along the fault (kinematic rupture). For spontaneous rupture, we examine the performance of the BDM by computing its convergence rate for an example in a homogeneous half-space with a slip-weakening friction law on the fault. By including the boundary integral representation on the fault, without using any refinement near the crack tips, the solution converges with the same rate as problems that do not exhibit any stress concentrations; that is, for the piecewise linear elements used here, the rms of the slip-rate distribution and of the traction distribution on the fault converge as O(Δx), whereas the slip distribution converges as O(Δx2), in which Δx is the mesh size of the finite element mesh. Additional 2-D spontaneous rupture simulations are performed for an inclined fault in a half-space for different values of the dip angle and for a folded layer system in a half-space, focusing only the P-SV case. The results show that the waves generated at the free surface and within the layers contribute to the propagation of the fault rupture and have a visible and sometimes strong effect on the ensuing ground motion.

In recent years, dynamic source models have been increasingly used to represent the rupture propa... more In recent years, dynamic source models have been increasingly used to represent the rupture propagation process during earthquakes. Several numerical approaches, such as the finite difference and the finite element methods, have been proposed for dynamic rupture modeling. These techniques, however, do not represent accurately stress fields close to the slip area, due to the presence of singularities at tips and edges. In practice, they can also have difficulty representing faults of arbitrary shape. The boundary integral equation method, on the other hand, can represent correctly the singular stress fields within the slip area and can deal with faults of arbitrary shape, but cannot be readily applied to complex media with heterogeneous materials and arbitrary geometries. In this study, we propose a new hybrid approach that combines the advantages of the boundary integral equation method and the domain finite element method. The formulation is presented in detail for the SH-wave problem, and is illustrated by several examples involving oblique faults in a highly heterogeneous half space with an irregular free surface.

The seismic rays from a given point source within a constant-gradient elastic isotropic medium (a... more The seismic rays from a given point source within a constant-gradient elastic isotropic medium (a medium with linear variation of wave velocities) are known to be circular. This feature comes along with the remarkable fact that the associated wave fronts are cylindrical or spherical in two or three-dimensions, respectively. The rays and wavefronts form a bipolar orthogonal system with well known properties. These unique circumstances allowed us to construct analytical approximations for the corresponding elastodynamic Green's functions. The derivation of our analytical approximations starts in 2D with the anti-plane SH line source making a generalization of the homogeneous medium Green's functions that relies on the asymptotic ray theory to establish both travel times and geometrical spreading factors. Our approximation accounts for both near-source effects and low frequencies. Moreover, the said correction is frequency-independent and this allows to obtain the Green's function in time domain. This remarkable result was heuristically extended to the in-plane P-SV line source vector case and was validated with the staggered stress-velocity pseudo spectral method with excellent results. Details are given in Sanchez-Sesma, Madariaga and Irikura (2001). In this paper our discovery is extended to 3D for a unit point force. Thus our approximation for the elastodynamic Green's function in a constant-gradient medium becomes the counterpart of the classical Stokes' problem for a homogeneous, isotropic elastic medium. The adequate behavior of our analytical expressions in 3D is tested by comparing them with results from a fourth order finite difference scheme. Acknowledgements This research have had the support of DGAPA-UNAM, Mexico, under grant IN121202 and CONACYT, Mexico, under project NC-204, by CICYT, Spain, under REN2002-04198-CO2-02/RIES, by the E.U. with FEDER, and the team RNM-194 of Junta de Andalucía, Spain. Reference Sanchez-Sesma, F. J., R. Madariaga and K. Irikura (2001), An approximate elastic 2D Green's functions for a constant gradient medium, Geophys. J. Int., 146, 237-248.

Geophysical Journal International, 2010

This paper presents a verification of three simulations of the ShakeOut scenario, an Mw 7.8 earth... more This paper presents a verification of three simulations of the ShakeOut scenario, an Mw 7.8 earthquake on a portion of the San Andreas fault in southern California, conducted by three different groups at the Southern California Earthquake Center using the SCEC Community Velocity Model for this region. We conducted two simulations using the finite difference method, and one by the finite element method, and performed qualitative and quantitative comparisons between the corresponding results. The results are in good agreement with each other; only small differences occur both in amplitude and phase between the various synthetics at ten observation points located near and away from the fault—as far as 150 km away from the fault. Using an available goodness-of-fit criterion all the comparisons scored above 8, with most above 9.2. This score would be regarded as excellent if the measurements were between recorded and synthetic seismograms. We also report results of comparisons based on time–frequency misfit criteria. Results from these two criteria can be used for calibrating the two methods for comparing seismograms. In those cases in which noticeable discrepancies occurred between the seismograms generated by the three groups, we found that they were the product of inherent characteristics of the various numerical methods used and their implementations. In particular, we found that the major source of discrepancy lies in the difference between mesh and grid representations of the same material model. Overall, however, even the largest differences in the synthetic seismograms are small. Thus, given the complexity of the simulations used in this verification, it appears that the three schemes are consistent, reliable and sufficiently accurate and robust for use in future large-scale simulations.

Soil Dynamics and Earthquake Engineering, 2009

... J. Alfonso Pérez-Ruiz a , b , E-mail The Corresponding Author , Leonardo Ramírez-Guzmán d , E... more ... J. Alfonso Pérez-Ruiz a , b , E-mail The Corresponding Author , Leonardo Ramírez-Guzmán d , E-mail The Corresponding Author and Andrés Pech e , E ... d Department of Civil and Environmental Engineering, Carnegie Mellon University, Forbes 5000 Pittsburgh, PA 15217, USA. ...

Parallel supercomputing has traditionally focused on the inner kernel of scientific simulations: ... more Parallel supercomputing has traditionally focused on the inner kernel of scientific simulations: the solver. The front and back ends of the simulation pipeline - problem description and interpretation of the output - have taken a back seat to the solver when it comes to attention paid to scalability and performance, and are often relegated to offline, sequential computation. As the largest simulations move beyond the realm of the terascale and into the petascale, this decomposition in tasks and platforms becomes increasingly untenable. We propose an end-to-end approach in which all simulation components - meshing, partitioning, solver, and visualization - are tightly coupled and execute in parallel with shared data structures and no intermediate I/O. We present our implementation of this new approach in the context of octree-based finite element simulation of earthquake ground motion. Performance evaluation on up to 2048 processors demonstrates the ability of the end-to-end approach to overcome the scalability bottlenecks of the traditional approach

We have developed a novel analytic capability for scientists and engineers to obtain insight from... more We have developed a novel analytic capability for scientists and engineers to obtain insight from ongoing large-scale parallel unstructured mesh simulations running on thousands of processors. The breakthrough is made possible by a new approach that visualizes partial differential equation (PDE) solution data simultaneously while a parallel PDE solver executes. The solution field is pipelined directly to volume rendering, which is computed in parallel using the same processors that solve the PDE equations. Because our approach avoids the bottlenecks associated with transferring and storing large volumes of output data, it offers a promising approach to overcoming the challenges of visualization of petascale simulations. The submitted video demonstrates real-time on-the-fly monitoring, interpreting, and steering from a remote laptop computer of a 1024processor simulation of the 1994 Northridge earthquake in Southern California.

State-of-the-art numerical solvers in Earth Sciences produce multi terabyte datasets per executio... more State-of-the-art numerical solvers in Earth Sciences produce multi terabyte datasets per execution. Operating on increasingly larger datasets becomes challenging due to insufficient data bandwidth. Queries result in difficult to handle I/O access patterns. BEMC is a new mechanism that allows querying and processing wavefields in the compressed representation. This approach combines well-known spatial-indexing techniques with novel compressed representations, thus reducing I/O bandwidth requirements. A new compression approach based on boundary integral representations exploits properties of the simulated domain. Frequency domain representation further compresses the data by eliminating temporal redundancy found in wave propagation data. This representation enables the transformation of a large I/O workload into a massively-parallel CPU-intensive computation. Queries to this representation result in largely sequential I/O accesses. Although, decompression places heavy demands on the CPU, it exhibits parallelism well-suited for many-core processors. We evaluate our approach in the context of data analysis for the Earth Sciences datasets.

The Wasatch Fault bounds the Salt Lake Basin (SLB), Utah to the east and is capable of producing ... more The Wasatch Fault bounds the Salt Lake Basin (SLB), Utah to the east and is capable of producing M7 earthquakes. In order to characterize the seismic hazard that such earthquakes pose to Salt Lake City and other heavily populated regions along the Wasatch Front, we require estimates of potential earthquake ground motions. However, the long recurrence time of these events (~1.3ky) necessitates that these estimates be derived from empirical relations observed in other regions or from deterministic modeling. Although previous and on-going studies (Liu et al., 2010; Roten et al., 2010) have begun to examine the earthquake ground motions caused by simulated M7 events on the Salt Lake Segment of the Wasatch Fault, a comprehensive validation of the Wasatch Community Velocity Model (WCVM) (Magistrale et al., 2006) has not yet been performed. We present the results from systematic testing of the intrinsic attenuation and seismic wave speed relations in the WCVM. Modifications to the empirical wavespeed relations and selection of intrinsic attenuation relations for the WCVM version 3c are made to minimize the differences between simulated and observed seismograms from small (M<5) earthquakes in the Wasatch Front. Simulations are carried out with the Hercules finite element tool-chain (Tu et al., 2006) and are valid at frequencies up to 1 Hz. We find a good agreement between body wave arrival times from the observed and simulated events. Discrepancies in the peak ground velocities and waveform durations are reduced in the final model. Some unresolved differences in the observed and synthetic seismograms persist and may result from un-modeled heterogeneities outside of the Salt Lake Basin. Future modeling efforts in the Wasatch Front may apply these results for improved earthquake ground motion estimates.

Abstract Conventional parallel scientific computing uses files as interface between simulation co... more Abstract Conventional parallel scientific computing uses files as interface between simulation components such as meshing, partitioning, solving and visualizing. This approach results in time-consuming file transfers, disk I/O and data format conversions ...

I present two numerical techniques to solve the Dynamic problem. I revisit and modify the Split N... more I present two numerical techniques to solve the Dynamic problem. I revisit and modify the Split Node approach and introduce a Stress Glut type Method. Both algorithms are implemented using a iso/sub- parametric FEM solver. In the first case, I discuss the formulation and perform an analysis of convergence for different orders of approximation for the acoustic case. I describe the algorithm of the second methodology as well as the assumptions made. The key to the new technique is to have an accurate representation of the traction. Thus, I devote part of the discussion to analyze the tractions for a simple example. The sensitivity of the method is tested by comparing against Split Node solutions.

Parallel supercomputing has traditionally focused on the inner kernel of scientific simulations: ... more Parallel supercomputing has traditionally focused on the inner kernel of scientific simulations: the solver. The front and back ends of the simulation pipeline---problem description and interpretation of the output---have taken a back seat to the solver when it comes to attention paid to scalability and performance, and are often relegated to offline, sequential computation. As the largest simulations move beyond

We have developed a novel analytic capability for scientists and engineers to obtain insight from... more We have developed a novel analytic capability for scientists and engineers to obtain insight from ongoing large-scale parallel unstructured mesh simulations running on thousands of processors. The breakthrough is made possible by a new approach that visualizes partial differential equation (PDE) solution data simultaneously while a parallel PDE solver executes. The solution field is pipelined directly to volume rendering, which is computed in parallel using the same processors that solve the PDE equations. Because our approach avoids the bottlenecks associated with transferring and storing large volumes of output data, it offers a promising approach to overcoming the challenges of visualization of petascale simulations. The submitted video demonstrates real-time on-the-fly monitoring, interpreting, and steering from a remote laptop computer of a 1024processor simulation of the 1994 Northridge earthquake in Southern California.

We demonstrate a new scalable approach to real-time monitoring, visualization, and steering of ma... more We demonstrate a new scalable approach to real-time monitoring, visualization, and steering of massively parallel simulations from a personal computer. The basis is an endto-end approach to parallel supercomputing in which all components -meshing, partitioning, solver, and visualization -are tightly coupled and execute in parallel on a supercomputer. This approach avoids bottlenecks associated with transfer and storage of massive simulation outputs, thereby enabling real-time visualization and steering on supercomputers with thousands of processors. We have incorporated this methodology into a framework named Hercules, which targets octree-based finite element simulations. The submitted video demonstrates real-time monitoring and steering from a laptop PC of a 1024-processor simulation of the 1994 Northridge earthquake in Southern California. Because this end-to-end approach does not require moving large data over the network and is completely scalable, our approach shows promise for overcoming the challenges of visualization of petascale simulations.

ABSTRACT Three-dimensional simulations of earthquake ground motion in sedimentary valleys have le... more ABSTRACT Three-dimensional simulations of earthquake ground motion in sedimentary valleys have led to a deeper understanding of wave propagation and site effects in urban regions. In this paper we present a preliminary study of the ground motion and resulting ...

Geophysical Journal International, 2010

We present a hybrid approach for solving the dynamic rupture problem in an anelastic medium. It c... more We present a hybrid approach for solving the dynamic rupture problem in an anelastic medium. It combines the advantages of the boundary integral equation method (BIEM), which is capable of representing accurately the solution near a crack tip, with those of the domain finite element method (FEM), which can handle conveniently heterogeneous materials, as well as the traction-free condition on a free surface and the continuity of traction across interfaces. When applied jointly, the proposed BIEM and the FEM can be used to solve efficiently spontaneous rupture problems in heterogeneous media, provided the rupture surface is contained within a homogeneous portion of the domain. The proposed method, BIEM-FEM approach (BDM, for Boundary/Domain Method), is verified for several antiplane (SH) and in-plane (P-SV) 2-D cases, in which the slip is prescribed along the fault (kinematic rupture). For spontaneous rupture, we examine the performance of the BDM by computing its convergence rate for an example in a homogeneous half-space with a slip-weakening friction law on the fault. By including the boundary integral representation on the fault, without using any refinement near the crack tips, the solution converges with the same rate as problems that do not exhibit any stress concentrations; that is, for the piecewise linear elements used here, the rms of the slip-rate distribution and of the traction distribution on the fault converge as O(Δx), whereas the slip distribution converges as O(Δx2), in which Δx is the mesh size of the finite element mesh. Additional 2-D spontaneous rupture simulations are performed for an inclined fault in a half-space for different values of the dip angle and for a folded layer system in a half-space, focusing only the P-SV case. The results show that the waves generated at the free surface and within the layers contribute to the propagation of the fault rupture and have a visible and sometimes strong effect on the ensuing ground motion.

In recent years, dynamic source models have been increasingly used to represent the rupture propa... more In recent years, dynamic source models have been increasingly used to represent the rupture propagation process during earthquakes. Several numerical approaches, such as the finite difference and the finite element methods, have been proposed for dynamic rupture modeling. These techniques, however, do not represent accurately stress fields close to the slip area, due to the presence of singularities at tips and edges. In practice, they can also have difficulty representing faults of arbitrary shape. The boundary integral equation method, on the other hand, can represent correctly the singular stress fields within the slip area and can deal with faults of arbitrary shape, but cannot be readily applied to complex media with heterogeneous materials and arbitrary geometries. In this study, we propose a new hybrid approach that combines the advantages of the boundary integral equation method and the domain finite element method. The formulation is presented in detail for the SH-wave problem, and is illustrated by several examples involving oblique faults in a highly heterogeneous half space with an irregular free surface.

The seismic rays from a given point source within a constant-gradient elastic isotropic medium (a... more The seismic rays from a given point source within a constant-gradient elastic isotropic medium (a medium with linear variation of wave velocities) are known to be circular. This feature comes along with the remarkable fact that the associated wave fronts are cylindrical or spherical in two or three-dimensions, respectively. The rays and wavefronts form a bipolar orthogonal system with well known properties. These unique circumstances allowed us to construct analytical approximations for the corresponding elastodynamic Green's functions. The derivation of our analytical approximations starts in 2D with the anti-plane SH line source making a generalization of the homogeneous medium Green's functions that relies on the asymptotic ray theory to establish both travel times and geometrical spreading factors. Our approximation accounts for both near-source effects and low frequencies. Moreover, the said correction is frequency-independent and this allows to obtain the Green's function in time domain. This remarkable result was heuristically extended to the in-plane P-SV line source vector case and was validated with the staggered stress-velocity pseudo spectral method with excellent results. Details are given in Sanchez-Sesma, Madariaga and Irikura (2001). In this paper our discovery is extended to 3D for a unit point force. Thus our approximation for the elastodynamic Green's function in a constant-gradient medium becomes the counterpart of the classical Stokes' problem for a homogeneous, isotropic elastic medium. The adequate behavior of our analytical expressions in 3D is tested by comparing them with results from a fourth order finite difference scheme. Acknowledgements This research have had the support of DGAPA-UNAM, Mexico, under grant IN121202 and CONACYT, Mexico, under project NC-204, by CICYT, Spain, under REN2002-04198-CO2-02/RIES, by the E.U. with FEDER, and the team RNM-194 of Junta de Andalucía, Spain. Reference Sanchez-Sesma, F. J., R. Madariaga and K. Irikura (2001), An approximate elastic 2D Green's functions for a constant gradient medium, Geophys. J. Int., 146, 237-248.

Geophysical Journal International, 2010

This paper presents a verification of three simulations of the ShakeOut scenario, an Mw 7.8 earth... more This paper presents a verification of three simulations of the ShakeOut scenario, an Mw 7.8 earthquake on a portion of the San Andreas fault in southern California, conducted by three different groups at the Southern California Earthquake Center using the SCEC Community Velocity Model for this region. We conducted two simulations using the finite difference method, and one by the finite element method, and performed qualitative and quantitative comparisons between the corresponding results. The results are in good agreement with each other; only small differences occur both in amplitude and phase between the various synthetics at ten observation points located near and away from the fault—as far as 150 km away from the fault. Using an available goodness-of-fit criterion all the comparisons scored above 8, with most above 9.2. This score would be regarded as excellent if the measurements were between recorded and synthetic seismograms. We also report results of comparisons based on time–frequency misfit criteria. Results from these two criteria can be used for calibrating the two methods for comparing seismograms. In those cases in which noticeable discrepancies occurred between the seismograms generated by the three groups, we found that they were the product of inherent characteristics of the various numerical methods used and their implementations. In particular, we found that the major source of discrepancy lies in the difference between mesh and grid representations of the same material model. Overall, however, even the largest differences in the synthetic seismograms are small. Thus, given the complexity of the simulations used in this verification, it appears that the three schemes are consistent, reliable and sufficiently accurate and robust for use in future large-scale simulations.

Soil Dynamics and Earthquake Engineering, 2009

... J. Alfonso Pérez-Ruiz a , b , E-mail The Corresponding Author , Leonardo Ramírez-Guzmán d , E... more ... J. Alfonso Pérez-Ruiz a , b , E-mail The Corresponding Author , Leonardo Ramírez-Guzmán d , E-mail The Corresponding Author and Andrés Pech e , E ... d Department of Civil and Environmental Engineering, Carnegie Mellon University, Forbes 5000 Pittsburgh, PA 15217, USA. ...

Parallel supercomputing has traditionally focused on the inner kernel of scientific simulations: ... more Parallel supercomputing has traditionally focused on the inner kernel of scientific simulations: the solver. The front and back ends of the simulation pipeline - problem description and interpretation of the output - have taken a back seat to the solver when it comes to attention paid to scalability and performance, and are often relegated to offline, sequential computation. As the largest simulations move beyond the realm of the terascale and into the petascale, this decomposition in tasks and platforms becomes increasingly untenable. We propose an end-to-end approach in which all simulation components - meshing, partitioning, solver, and visualization - are tightly coupled and execute in parallel with shared data structures and no intermediate I/O. We present our implementation of this new approach in the context of octree-based finite element simulation of earthquake ground motion. Performance evaluation on up to 2048 processors demonstrates the ability of the end-to-end approach to overcome the scalability bottlenecks of the traditional approach

We have developed a novel analytic capability for scientists and engineers to obtain insight from... more We have developed a novel analytic capability for scientists and engineers to obtain insight from ongoing large-scale parallel unstructured mesh simulations running on thousands of processors. The breakthrough is made possible by a new approach that visualizes partial differential equation (PDE) solution data simultaneously while a parallel PDE solver executes. The solution field is pipelined directly to volume rendering, which is computed in parallel using the same processors that solve the PDE equations. Because our approach avoids the bottlenecks associated with transferring and storing large volumes of output data, it offers a promising approach to overcoming the challenges of visualization of petascale simulations. The submitted video demonstrates real-time on-the-fly monitoring, interpreting, and steering from a remote laptop computer of a 1024processor simulation of the 1994 Northridge earthquake in Southern California.

State-of-the-art numerical solvers in Earth Sciences produce multi terabyte datasets per executio... more State-of-the-art numerical solvers in Earth Sciences produce multi terabyte datasets per execution. Operating on increasingly larger datasets becomes challenging due to insufficient data bandwidth. Queries result in difficult to handle I/O access patterns. BEMC is a new mechanism that allows querying and processing wavefields in the compressed representation. This approach combines well-known spatial-indexing techniques with novel compressed representations, thus reducing I/O bandwidth requirements. A new compression approach based on boundary integral representations exploits properties of the simulated domain. Frequency domain representation further compresses the data by eliminating temporal redundancy found in wave propagation data. This representation enables the transformation of a large I/O workload into a massively-parallel CPU-intensive computation. Queries to this representation result in largely sequential I/O accesses. Although, decompression places heavy demands on the CPU, it exhibits parallelism well-suited for many-core processors. We evaluate our approach in the context of data analysis for the Earth Sciences datasets.