Search NASA⌕ Search

SEARCH · Search NASA

Results for “parallel computer architecture”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,621 records · Page 90

Toward Exascale: Overview of Large Eddy Simulations and Direct Numerical Simulations of Nuclear Reactor Flows with the Spectral Element Method in Nek5000

At the beginning of the last decade, Petascale supercomputers (i.e., computers capable of more than 1 petaFLOP) emerged. Now, at the dawn of exascale supercomputing, we provide a review of recent landmark simulations of portions of reactor components with turbulence-resolving techniques that this computational power has made possible. In fact, these simulations have provided invaluable insight into flow dynamics, which is difficult or often impossible to obtain with experiments alone. We focus on simulations performed with the spectral element method, as this method has emerged as a powerful tool to deliver massively parallel calculations at high fidelity by using large eddy simulation or direct numerical simulation. We also limit this paper to constant-property incompressible flow of a Newtonian fluid in the absence of other body or external forces, although the method is by no means limited to this class of flows. We briefly review the fundamentals of the method and the reasons it is compelling for the simulation of nuclear engineering flows. We review in detail a series of Petascale simulations, including the simulations of helical coil steam generators, fuel assemblies, and pebble beds. Even with Petascale computing, however, limitations for nuclear modeling and simulation tools remain. In particular, the size and scope of turbulence-resolving simulations are still limited by computing power and resolution requirements, which scale with the Reynolds number. In the final part of this paper, we discuss the future of the field, including recent advancements in emerging architectures such as GPUbased supercomputers, which are expected to power the next generation of high-performance computers.

computational fluid dynamics↗

Affinity-aware checkpoint restart

Current checkpointing techniques employed to overcome faults for HPC applications result in inferior application performance after restart from a checkpoint for a number of applications. This is due to a lack of page and core affinity awareness of the checkpoint/restart (C/R) mechanism, i.e., application tasks originally pinned to cores may be restarted on different cores, and in case of non-uniform memory architectures (NUMA), quite common today, memory pages associated with tasks on a NUMA node may be associated with a different NUMA node after restart. Here, this work contributes a novel design technique for C/R mechanisms to preserve task-to-core maps and NUMA node specific page affinities across restarts. Experimental results with BLCR, a C/R mechanism, enhanced with affinity awareness demonstrate significant performance benefits of 37%-73% for the NAS Parallel Benchmark codes and 6-12% for NAMD with negligible overheads instead of up to nearly four times longer an execution times without affinity-aware restarts on 16 cores.

97 MATHEMATICS AND COMPUTING↗

Scalable Memory Registration for High-Performance Networks Using Helper Threads

Remote DMA (RDMA) enables high performance networks to reduce data copying between an application and the operating system (OS). However RDMA operations in some high performance networks require communication memory explicitly registered with the network adapter and pinned by the OS. Memory registration and pinning limits the flexibility of the memory system and reduces the amount of memory that user processes can allocate. These issues become more significant on multicore platforms, since registered memory demand grows linearly with the number of processor cores. In this paper we propose a new memory registration/deregistration strategy to reduce registered memory on multicore architectures for HPC applications. We hide the cost of dynamic memory management by offloading all dynamic memory registration and deregistration requests to a dedicated memory management helper thread. We investigate design policies and performance implications of the helper thread approach. We evaluate our framework with the NAS parallel benchmarks, for which our registration scheme significantly reduces the registered memory (23.62% on average and up to 49.39%) and avoids memory registration/deregistration costs for reused communication memory. We show that our system enables the execution of problem sizes that could not complete under existing memory registration strategies.

99 GENERAL AND MISCELLANEOUS↗

A nonvoxel-based dose convolution/superposition algorithm optimized for scalable GPU architectures

Purpose: Real-time adaptive planning and treatment has been infeasible due in part to its high computational complexity. There have been many recent efforts to utilize graphics processing units (GPUs) to accelerate the computational performance and dose accuracy in radiation therapy. Data structure and memory access patterns are the key GPU factors that determine the computational performance and accuracy. In this paper, the authors present a nonvoxel-based (NVB) approach to maximize computational and memory access efficiency and throughput on the GPU. Methods: The proposed algorithm employs a ray-tracing mechanism to restructure the 3D data sets computed from the CT anatomy into a nonvoxel-based framework. In a process that takes only a few milliseconds of computing time, the algorithm restructured the data sets by ray-tracing through precalculated CT volumes to realign the coordinate system along the convolution direction, as defined by zenithal and azimuthal angles. During the ray-tracing step, the data were resampled according to radial sampling and parallel ray-spacing parameters making the algorithm independent of the original CT resolution. The nonvoxel-based algorithm presented in this paper also demonstrated a trade-off in computational performance and dose accuracy for different coordinate system configurations. In order to find the best balance between the computed speedup and the accuracy, the authors employed an exhaustive parameter search on all sampling parameters that defined the coordinate system configuration: zenithal, azimuthal, and radial sampling of the convolution algorithm, as well as the parallel ray spacing during ray tracing. The angular sampling parameters were varied between 4 and 48 discrete angles, while both radial sampling and parallel ray spacing were varied from 0.5 to 10 mm. The gamma distribution analysis method (γ) was used to compare the dose distributions using 2% and 2 mm dose difference and distance-to-agreement criteria, respectively. Accuracy was investigated using three distinct phantoms with varied geometries and heterogeneities and on a series of 14 segmented lung CT data sets. Performance gains were calculated using three 256 mm cube homogenous water phantoms, with isotropic voxel dimensions of 1, 2, and 4 mm. Results: The nonvoxel-based GPU algorithm was independent of the data size and provided significant computational gains over the CPU algorithm for large CT data sizes. The parameter search analysis also showed that the ray combination of 8 zenithal and 8 azimuthal angles along with 1 mm radial sampling and 2 mm parallel ray spacing maintained dose accuracy with greater than 99% of voxels passing the γ test. Combining the acceleration obtained from GPU parallelization with the sampling optimization, the authors achieved a total performance improvement factor of >175 000 when compared to our voxel-based ground truth CPU benchmark and a factor of 20 compared with a voxel-based GPU dose convolution method. Conclusions: The nonvoxel-based convolution method yielded substantial performance improvements over a generic GPU implementation, while maintaining accuracy as compared to a CPU computed ground truth dose distribution. Such an algorithm can be a key contribution toward developing tools for adaptive radiation therapy systems.

60 APPLIED LIFE SCIENCES↗

Index Sets and Vectorization

Vectorization is data parallelism (SIMD, SIMT, etc.) - extension of ISA enabling the same instruction to be performed on multiple data items simultaeously. Many/most CPUs support vectorization in some form. Vectorization is difficult to enable, but can yield large efficiency gains. Extra programmer effort is required because: (1) not all algorithms can be vectorized (regular algorithm structure and fine-grain parallelism must be used); (2) most CPUs have data alignment restrictions for load/store operations (obey or risk incorrect code); (3) special directives are often needed to enable vectorization; and (4) vector instructions are architecture-specific. Vectorization is the best way to optimize for power and performance due to reduced clock cycles. When data is organized properly, a vector load instruction (i.e. movaps) can replace 'normal' load instructions (i.e. movsd). Vector operations can potentially have a smaller footprint in the instruction cache when fewer instructions need to be executed. Hybrid index sets insulate users from architecture specific details. We have applied hybrid index sets to achieve optimal vectorization. We can extend this concept to handle other programming models.

97 MATHEMATICS AND COMPUTING↗

Evolution of the SLATE linear algebra library

SLATE (Software for Linear Algebra Targeting Exascale) is a distributed, dense linear algebra library targeting both CPU-only and GPU-accelerated systems, developed over the course of the Exascale Computing Project (ECP). While it began with several documents setting out its initial design, significant design changes occurred throughout its development. In some cases, these were anticipated: an early version used a simple consistency flag that was later replaced with a full-featured consistency protocol. In other cases, performance limitations and software and hardware changes prompted a redesign. Sequential communication tasks were parallelized; host-to-host MPI calls were replaced with GPU device-to-device MPI calls; more advanced algorithms such as Communication Avoiding LU and the Random Butterfly Transform (RBT) were introduced. Early choices that turned out to be cumbersome, error prone, or inflexible have been replaced with simpler, more intuitive, or more flexible designs. Applications have been a driving force, prompting a lighter weight queue class, nonuniform tile sizes, and more flexible MPI process grids. Of paramount importance has been building a portable library that works across several different GPU architectures – AMD, Intel, and NVIDIA – while keeping a clean and maintainable codebase. Here we explore the evolving design choices and their effects, both in terms of performance and software sustainability.

Gates, Mark↗

Message Passing vs. Shared Address Space on a Cluster of SMPs

The convergence of scalable computer architectures using clusters of PCs (or PC-SMPs) with commodity networking has become an attractive platform for high end scientific computing. Currently, message-passing and shared address space (SAS) are the two leading programming paradigms for these systems. Message-passing has been standardized with MPI, and is the most common and mature programming approach. However message-passing code development can be extremely difficult, especially for irregular structured computations. SAS offers substantial ease of programming, but may suffer from performance limitations due to poor spatial locality, and high protocol overhead. In this paper, we compare the performance of and programming effort, required for six applications under both programming models on a 32 CPU PC-SMP cluster. Our application suite consists of codes that typically do not exhibit high efficiency under shared memory programming. due to their high communication to computation ratios and complex communication patterns. Results indicate that SAS can achieve about half the parallel efficiency of MPI for most of our applications: however, on certain classes of problems SAS performance is competitive with MPI. We also present new algorithms for improving the PC cluster performance of MPI collective operations.

Shan, Hongzhang↗

Comparing High Performance Computing Accelerator Programming Models

Accelerator devices are becoming a norm in High Performance Computing (HPC). With more systems opting for heterogeneous architectures, portable programming models like OpenMP and OpenACC are becoming increasingly important. The SPEC ACCEL 1.2 benchmark suite consists of comparable benchmarks in OpenCL, OpenMP 4.5, and OpenACC 2.5 that can be used to evaluate the performance and support for programming models and frameworks on heterogeneous platforms. In this paper we go beneath the normative metric of performance times and look at the individual kernels to study the usage, strengths, and weaknesses of the two prevalent portable heterogeneous programming models, OpenMP and OpenACC. From our analysis we identify that benchmarks like MRI-Q, SP and BT have better performance using OpenACC, while benchmarks like MiniGhost, LBM and LBDC do consistently better with the OpenMP programming model across supercomputers like Titan, and Summit.We deep dive into the kernels of select four benchmarks to answer questions like: Where does the benchmark spend most of its cycles? What is the parallelization strategy used? Why is one programming model more performant than the other? By identifying the similarities and dierences we want to contrast between the benchmark implementation strategies in the SPEC ACCEL 1.2 benchmarks and provide more insights into the OpenMP and OpenACC programming models.

Pophale, Swaroop S.↗

Modeling Data Movement Performance on Heterogeneous Architectures

The cost of data movement on parallel systems varies greatly with machine architecture, job partition, and nearby jobs. Performance models that accurately capture the cost of data movement provide a tool for analysis, allowing for communication bottlenecks to be pinpointed. Modern heterogeneous architectures yield increased variance in data movement as there are a number of viable paths for inter-GPU communication. In this paper, we present performance models for the various paths of inter-node communication on modern heterogeneous architectures, including the trade-off between GPUDirect communication and copying to CPUs. Furthermore, we present a novel optimization for inter-node communication based on these models, utilizing all available CPU cores per node. Finally, we show associated performance improvements for MPI collective operations.

97 MATHEMATICS AND COMPUTING↗

MemXCT: Design, Optimization, Scaling, and Reproducibility of X-Ray Tomography Imaging

Here, this work extends our previous research entitled "MemXCT: Memory-centric X-ray CT Reconstruction with Massive Parallelization" that was originally published at SC19 conference (Hidayetoglu et al., 2019) with reproducibility of the computational imaging performance. X-ray computed tomography (XCT) is regularly used at synchrotron light sources to study the internal morphology of materials at high resolution. However, experimental constraints, such as radiation sensitivity, can result in noisy or undersampled measurements. Further, depending on the resolution, sample size and data acquisition rates, the resulting noisy dataset can be in the order of terabytes. Advanced iterative reconstruction techniques can produce high-quality images from noisy measurements, but their computational requirements have made their use an exception rather than the rule. We propose a novel memory-centric approach that avoids redundant computations at the expense of additional memory complexity. We develop a memory-centric iterative reconstruction system, MemXCT, that uses an optimized SpMV implementation with two-level pseudo-Hilbert ordering and multi-stage input buffering. We evaluate MemXCT on various supercomputer architectures involving KNL and GPU. MemXCT can reconstruct a large (11Kx11K) mouse brain tomogram in 10 seconds using 4096 KNL nodes (256K cores). The results presented in our original article at the SC19 were based on large-scale supercomputing resources. The MemXCT application was selected for the Student Cluster Competition (SCC) Reproducibility Challenge and evaluated on a variety of cloud computing resources by universities around the world in the SC20 conference. We summarize the results of the top-ranked SCC Reproducibility Challenge teams and identify the most pertinent measures for ensuring the reproducibility of our experiments in this article.

47 OTHER INSTRUMENTATION↗

Adding GPU Support to the Markov Chain Monte Carlo Code Catmip

In geophysics, we are confronted with many under-determined inverse problems. For example, all of our observations of earthquakes are made at the Earth’s surface. So, when we try to infer how slip during an earthquake evolves in space and time, we find that there are many potential slip histories that are consistent with our limited observations and our understanding of earthquake physics. One way to approach these problems is with Bayesian analysis which allows us to infer the ensemble of all potential slip models that satisfy the observations and our prior knowledge of earthquake physics. In Bayesian analysis, our prior knowledge is known as the prior probability density function or prior PDF, the fit to the data is known as the data likelihood, and the target PDF that satisfies both the prior PDF and data likelihood is known as the posterior PDF. However, simulating the posterior PDF typically requires using Markov Chain Monte Carlo (MCMC) to draw tens of billions of random realizations of earthquake slip models, which may not be computationally feasible. To make this and similar geophysical inversions computationally tractable, we developed the Cascading Adaptive Transitional Metropolis In Parallel (CATMIP) algorithm. CATMIP is an efficient parallel Markov Chain Monte Carlo (MCMC) sampler that is used for model fitting and uncertainty quantification in geophysics. Example use cases are earthquake rupture modeling, determining mineral composition on Mars, reconstructing the history of ocean salinity, and historical earthquake relocation. CATMIP employs many parallel instances of the Metropolis algorithm for sampling in a transitioning framework. Transitioning is a process in which a set of random samples at equilibrium with a known probability density function (PDF) are used as seeds for the Markov chains to sample successive target PDFs that incrementally move the distribution from the starting seeds to the final desired PDF that describes the relative plausibility of potential values for the model parameters. The algorithm is implemented as a Master-Worker model employing MPI for communication. The worker processes are loosely coupled with global parameters periodically optimized by the master process. This provides a very high amount of parallelism with little communication between updates. During the presentation we will discuss the history of the algorithm and elaborate the earthquake rupture modeling use case for the CATMIP package. Our first step toward GPU optimization was to optimize the code for the CPU. CPU profiling revealed that most of the compute time is spent in calls to level 2 BLAS routines and calls to GSL random number generators. We revised the algorithm to employ level 3 BLAS routines instead. In our presentation we will describe how this was accomplished. Adding GPU support to CATMIP consisted mostly of replacing the calls to GSL with calls to GPU vendor-provided library routines. A small number of loops were directly implemented in CUDA. In the presentation will provide implementation details. Finally, we will discuss methods for profiling and opportunities for further optimizing GPU execution. By creating a code with the flexibility to run on either a CPU or GPU architecture, CATMIP can be used on systems ranging from large CPU-based HPC environments to single servers with GPU acceleration and everything in between.

HECC↗

Editorial: Neuroscience, computing, performance, and benchmarks: Why it matters to neuroscience how fast we can compute

At the turn of the millennium the computational neuroscience community realized that neuroscience was in a software crisis: software development was no longer progressing as expected and reproducibility declined. The International Neuroinformatics Coordinating Facility (INCF) was inaugurated in 2007 as an initiative to improve this situation. The INCF has since pursued its mission to help the development of standards and best practices. In a community paper published this very same year, Brette et al. tried to assess the state of the field and to establish a scientific approach to simulation technology, addressing foundational topics, such as which simulation schemes are best suited for the types of models we see in neuroscience. In 2015, a Frontiers Research Topic “Python in neuroscience” by Muller et al. triggered and documented a revolution in the neuroscience community, namely in the usage of the scripting language Python as a common language for interfacing with simulation codes and connecting between applications. The review by Einevoll et al. documented that simulation tools have since further matured and become reliable research instruments used by many scientific groups for their respective questions. Open source and community standard simulators today allow research groups to focus on their scientific questions and leave the details of the computational work to the community of simulator developers. A parallel development has occurred, which has been barely visible in neuroscientific circles beyond the community of simulator developers: Supercomputers used for large and complex scientific calculations have increased their performance from ~10 TeraFLOPS (10 13 floating point operations per second) in the early 2000s to above 1 ExaFLOPS (10 18 floating point operations per second) in the year 2022. This represents a 100,000-fold increase in our computational capabilities, or almost 17 doublings of computational capability in 22 years. Moore's law (the observation that it is economically viable to double the number of transistors in an integrated circuit every other 18–24 months) explains a part of this; our ability and willingness to build and operate physically larger computers, explains another part. It should be clear, however, that such a technological advancement requires software adaptations and under the hood, simulators had to reinvent themselves and change substantially to embrace this technological opportunity. It actually is quite remarkable that—apart from the change in semantics for the parallelization—this has mostly happened without the users knowing. The current Research Topic was motivated by the wish to assemble an update on the state of neuroscientific software (mostly simulators) in 2022, to assess whether we can see more clearly which scientific questions can (or cannot) be asked due to our increased capability of simulation, and also to anticipate whether and for how long we can expect this increase of computational capabilities to continue.

biophysically detailed models↗

Progress In Developing An In-Pile Acoustically Telemetered Sensor Infrastructure

A salient grand challenge for a number of Department of Energy programs such as Fuels Cycle Research and Development ( includes Accident Tolerant Fuel research and the Transient Reactor Test Facility Restart experiments), Light Water Sustainability, and Advanced Reactor Technologies is to enhance our fundamental understanding of fuel and materials behavior under irradiation. Robust and accurate in-pile measurements will be instrumental to develop and validate a computationally predictive multi-scale understanding of nuclear fuel and materials. This sensing technology will enable the linking of fundamental micro-structural evolution mechanisms to the macroscopic degradation of fuels and materials. The in situ sensors and measurement systems will monitor local environmental parameters as well as characterize microstructure evolution during irradiation. One of the major road blocks in developing practical robust, and cost effective in-pile sensor systems, are instrument leads. If a wireless telemetry infrastructure can be developed for in-pile use, in-core measurements would become more attractive and effective. Thus to be successful in accomplishing effective in-pile sensing and microstructure characterization an interdisciplinary measurement infrastructure needs to be developed in parallel with key sensing technology. For the discussion in this research, infrastructure is defined as systems, technology, techniques, and algorithms that may be necessary in the delivery of beneficial and robust data from in-pile devices. The architecture of a system’s infrastructure determines how well it operates and how flexible it is to meet future requirements. The limiting path for the effective deployment of the salient sensing technology will not be the sensors themselves but the infrastructure that is necessary to communicate data from in-pile to the outside world in a non-intrusive and reliable manner. This article gives a high level overview of a promising telemetry infrastructure based on acoustic wireless transmission of data that is being developed and tested by the INL, Penn State and Westinghouse.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

A Navier-Strokes Chimera Code on the Connection Machine CM-5: Design and Performance

We have implemented a three-dimensional compressible Navier-Stokes code on the Connection Machine CM-5. The code is set up for implicit time-stepping on single or multiple structured grids. For multiple grids and geometrically complex problems, we follow the 'chimera' approach, where flow data on one zone is interpolated onto another in the region of overlap. We will describe our design philosophy and give some timing results for the current code. A parallel machine like the CM-5 is well-suited for finite-difference methods on structured grids. The regular pattern of connections of a structured mesh maps well onto the architecture of the machine. So the first design choice, finite differences on a structured mesh, is natural. We use centered differences in space, with added artificial dissipation terms. When numerically solving the Navier-Stokes equations, there are liable to be some mesh cells near a solid body that are small in at least one direction. This mesh cell geometry can impose a very severe CFL (Courant-Friedrichs-Lewy) condition on the time step for explicit time-stepping methods. Thus, though explicit time-stepping is well-suited to the architecture of the machine, we have adopted implicit time-stepping. We have further taken the approximate factorization approach. This creates the need to solve large banded linear systems and creates the first possible barrier to an efficient algorithm. To overcome this first possible barrier we have considered two options. The first is just to solve the banded linear systems with data spread over the whole machine, using whatever fast method is available. This option is adequate for solving scalar tridiagonal systems, but for scalar pentadiagonal or block tridiagonal systems it is somewhat slower than desired. The second option is to 'transpose' the flow and geometry variables as part of the time-stepping process: Start with x-lines of data in-processor. Form explicit terms in x, then transpose so y-lines of data are in-processor. Form explicit terms in y, then transpose so z-lines are in processor. Form explicit terms in z, then solve linear systems in the z-direction. Transpose to the y-direction, then solve linear systems in the y-direction. Finally transpose to the x direction and solve linear systems in the x-direction. This strategy avoids inter-processor communication when differencing and solving linear systems, but requires a large amount of communication when doing the transposes. The transpose method is more efficient than the non-transpose strategy when dealing with scalar pentadiagonal or block tridiagonal systems. For handling geometrically complex problems the chimera strategy was adopted. For multiple zone cases we compute on each zone sequentially (using the whole parallel machine), then send the chimera interpolation data to a distributed data structure (array) laid out over the whole machine. This information transfer implies an irregular communication pattern, and is the second possible barrier to an efficient algorithm. We have implemented these ideas on the CM-5 using CMF (Connection Machine Fortran), a data parallel language which combines elements of Fortran 90 and certain extensions, and which bears a strong similarity to High Performance Fortran. We make use of the Connection Machine Scientific Software Library (CMSSL) for the linear solver and array transpose operations.

Jespersen, Dennis C.↗

Scaling soft matter physics to thousands of graphics processing units in parallel

We describe a multi-graphics processing unit (GPU) implementation of the Ludwig application, which specialises in simulating a variety of complex fluids via lattice Boltzmann fluid dynamics coupled to additional physics describing complex fluid constituents. We describe our methodology in augmenting the original central processing unit (CPU) version with GPU functionality in a maintainable fashion. We present several optimisations that maximise performance on the GPU architecture through tuning for the GPU memory hierarchy. We describe how we implement particles within the fluid in such a way to avoid a major diversion of the CPU and GPU codebases, whilst minimising data transfer at each time step. We detail our halo-exchange communication phase for the code, which exploits overlapping to allow efficient parallel scaling to many GPUs. We present results showing that the application demonstrates excellent scaling to at least 8192 GPUs in parallel, the largest system tested at the time of writing. The GPU version (on NVIDIA K20X GPUs) is around 3.5–5 times faster that the CPU version (on fully utilised AMD Opteron 6274 16-core CPUs), comparing equal numbers of CPUs and GPUs.

Computer Science↗

Accuracy of the explicit energy-conserving particle-in-cell method for under-resolved simulations of capacitively coupled plasma discharges

The traditional explicit electrostatic momentum-conserving particle-in-cell algorithm requires strict resolution of the electron Debye length to deliver numerical stability and accuracy. The explicit electrostatic energy-conserving particle-in-cell algorithm alleviates this constraint with minimal modification to the traditional algorithm, retaining its simplicity, ease of parallelization, and acceleration on modern supercomputing architectures. In this article, we apply the algorithm to model a one-dimensional radio frequency capacitively coupled plasma discharge relevant to industrial applications. The energy-conserving approach closely matches the results from the momentum-conserving algorithm and retains accuracy even for cell sizes up to 8 times the electron Debye length. For even larger cells, the algorithm loses accuracy due to poor resolution of steep gradients within the radio frequency sheath. Accuracy can be recovered by adopting a non-uniform grid, which resolves the sheath and allows for cell sizes up to 32 times the electron Debye length in the quasi-neutral bulk of the discharge. The effect is an up to 8 times reduction in the number of required simulation cells, an improvement that can compound in higher-dimensional simulations. We therefore consider the explicit energy-conserving algorithm as a promising approach to significantly reduce the computational cost of full-scale device simulations and a pathway to delivering kinetic simulation capabilities of use to industry.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

How Much Chemistry Does a Deep Neural Network Need to Know to Make Accurate Predictions?

In the last few years, we have seen the rise of deep learning applications in a broad range of computational chemistry research problems. Using human-engineered chemical features, such as molecular descriptors and fingerprints, deep learning models have shown similar, if not better performance that most traditional machine learning algorithms. Recently, we reported on the development of Chemception, a deep convolutional neural network (CNN) architecture for general-purpose small molecule property prediction. On average, Chemception matched the performance of expert-developed QSAR/QSPR models trained on chemical features (molecular fingerprints), despite that it was trained on just 2D images of molecular drawings with minimal chemical information. Here, we investigate the effects of systematically removing and adding basic chemical information to the image channels of the 2D images used to train Chemception. By augmenting our images with only 3 additional basic chemical information, we demonstrate the improvement of Chemception performance – that it is now more accurate than contemporary deep learning models trained on ECFP fingerprints for the prediction of toxicity, activity and solvation free energy, as well as physics-based free energy simulation methods for computing solvation properties. By altering the chemical information content in the image channels, and examining the resulting performance of Chemception, we also identify to two different “learning patterns” in toxicity/activity as compared to solvation free energy, and it parallels the fundamental differences in contemporary chemistry research for predicting toxicity/activity and solvation free energy.

Artificial Intelligence, Cheminformatics, Computat↗

Understanding Aprun Use Patterns

On the Cray XT, aprun is the command to launch an application to a set of compute nodes reserved through the Application Level Placement Scheduler (ALPS). At the National Energy Research Scientific Computing Center (NERSC), interactive aprun is disabled. That is, invocations of aprun have to go through the batch system. Batch scripts can and often do contain several apruns which either use subsets of the reserved nodes in parallel, or use all reserved nodes in consecutive apruns. In order to better understand how NERSC users run on the XT, it is necessary to associate aprun information with jobs. It is surprisingly more challenging than it sounds. In this paper, we describe those challenges and how we solved them to produce daily per-job reports for completed apruns. We also describe additional uses of the data, e.g. adjusting charging policy accordingly or associating node failures with jobs/users, and plans for enhancements.

97 MATHEMATICS AND COMPUTING↗