CyberShake is a computational platform developed by the Southern California Earthquake Center (SCEC) that explicitly incorporates earthquake rupture time histories and deterministic wave propagation effects into seismic hazard calculations through the use of 3D waveform simulations. Using CyberShake, SCEC has created the first physics-based probabilistic seismic hazard analysis (PSHA) models of the Los Angeles region from suites of simulations comprising ~10 8 seismograms. The current models are, however, limited to low seismic frequencies (≤ 0.5 Hz). To increase the maximum simulated frequency to above 1 Hz and produce a California state-wide model, we have transformed SCEC Anelastic Wave Propagation code (AWP-ODC) to include strain Green's tensor (SGT) calculations to accelerate CyberShake calculations. This tensor-valued wave field code has both CPU and GPU components in place for flexibility on different architectures. We have demonstrated the performance and scalability of this solver optimized for the heterogeneous Blue Waters system at NCSA. The high performance of the wave propagation computation, coupled with CPU/GPU co-scheduling capabilities of our workflow-managed systems, make a statewide hazard model a goal reachable with existing supercomputers.

We use an efficient earthquake simulator that incorporates rate-state constitutive properties and uses boundary element method to discretize the fault surfaces, to generate the synthetic earthquakes in the fault system. Rate-and-state seismicity equation is subsequently employed to calculate the seismicity rate in a region of interest using the Coulomb stress transfer from the main shocks in the fault system. The Coulomb stress transfer is obtained by resolving the induced stresses due to the fault patch slips onto the optimal-oriented fault planes. The example results show that immediately after a main shock the aftershocks are concentrated in the vicinity of the rupture area due to positive stress transfers and then disperse away into the surrounding region toward the background rate distribution. The number of aftershocks near the rupture region is found to decay with time as Omori aftershock decay law predicts. The example results demonstrate that the rate-and-state fault system earthquake simulator and the seismicity equations based on the rate-state friction nucleation of earthquake are well posited to characterize the aftershock distribution in regional assessments of earthquake probabilities.

Data-intensive science frontiers and challenges are emerg ing as computer technology has evolved substantially. Large-scale simu lations demand significant I/O workload, and as a result the I/O performance often becomes a bottleneck preventing high performance in scientific applications. In this paper we introduce a variety of I/O optimization techniques developed and imple mented when scaling a seismic application to petascale. These techniques include file system striping, data aggreg ation, reader/writer limit ing and less interleaving of data, collective MPI-IO, and data staging. The optimizations result in nearly perfect scalability of the target applicat ion on some of the most advanced petascale systems. The techniques introduced in this paper are applicab le to other scientific applications facing similar petascale I/O challenges.

We have developed a highly scalable 3D Finite Difference GPU code for use in earthquake engineering and disaster management through regional petascale earthquake simulations. This MPI-CUDA code is based on a widely-used wave propagation code called AWP-ODC and restructured for high throughput and efficiency on a heterogeneous computing architecture. We present an effective communication reduction technique for leveraging GPUs with minimal PCI-e overhead, and a novel overlapping method to fully hide data communication latency between GPUs. The optimization concept used in this work can be extended to general stencil computing on a structured grid. The benchmarks demonstrated sustained 100 TFlops in single precision for 49 billion mesh points using 952 GPUs on the NCCS Titan Phase 5 system, which is a 77-fold speedup compared to the CPU version of the code. This multi-GPU implementation has been validated and used for a large-scale verification wave propagation simulation of Mw5.4 Chino Hills earthquake using 128 GPUs.

We have developed a highly scalable and efficient GPU-based finite-difference code (AWP) for earthquake simulation that implements high throughput, memory locality, communication reduction and communication / computation overlap and achieves linear scalability on Cray XK7 Titan at ORNL and NCSA's Blue Waters system. We simulate realistic 0-10 Hz earthquake ground motions relevant to building engineering design using high-performance AWP. Moreover, we show that AWP provides a speedup by a factor of 110 in key strain tensor calculations critical to probabilistic seismic hazard analysis (PSHA). These performance improvements to critical scientific application software, coupled with improved co-scheduling capabilities of our workflow-managed systems, make a statewide hazard model a goal reachable with existing supercomputers. The performance improvements of GPU-based AWP are expected to save millions of core-hours over the next few years as physics-based seismic hazard analysis is developed using heterogeneous petascale supercomputers.

We propose a general query-based framework of an information exchange system for information search and acquisition. The system is intended to support ad-hoc and potentially anonymous interactions between users of the network, while apparently providing a convenient and effective mechanism to acquire information of the “most interest.” In particular, members of the population may cooperate over a network to reliably identify and acquire the most popular “answers” to the most popular “questions,” where popularity is defined in a global sense by summing up rankings over all users. We propose a fully distributed algorithm for reliable identification of the most popular content, which attempts to minimize and balance the amount of content that needs to be examined and ranked by each user. Using a Zipf product form model for the preferences of a random user in the system, we present simulation results which suggest that the volume of content that needs to be examined by each user in the system, in order to reliably identify the k-most popular items, is effectively and approximately O(3k).

We consider a problem that is motivated by the problem of efficiently determining the opinions of a large number of people, where some fraction of the population may be untrustworthy. We propose a class of distributed algorithms called iterative majority voting (IMV), which appears to solve the problem efficiently as long as a positive fraction of the population is trustworthy. IMV consists of two phases: sampling and networking. In the sampling phase each person samples some random set of people for their opinions. In the networking phase each person communicates to random people about their estimates on the group opinion, and the approach benefits from “the power of networking.” We show that IMV is a fair, efficient and robust distributed algorithm by theoretical calculations and simulations. Possible applications include distributed ranking systems, survey and polling systems, social contexts in systems with or without an infrastructure (wireless ad-hoc networks or sensor networks).

Polar codes are a recently introduced class of codes that achieve the capacity of arbitrary symmetric binary-input channels. This capacity-achieving performance is obtained by encoders and decoders of complexity O(N log N) where N is the code block-length. The performance of polar coding under belief propahation (BP) decoding has been studied before, using Reed-Muller (RM) codes as a benchmark. This work studies the performance of polar coding under trellis-based maximum-likelihood (ML) decoding, again using RM codes as a benchmark. One finding is that RM codes perform better than polar codes under ML decoding for certain short codes. On the other hand, polar codes have a lower trellis complexity. A second finding is that BP decoding offers performance comparable to ML decoding as the block-length is increased.