Search NASA⌕ Search

SEARCH · Search NASA

Results for “Cluster Computing”

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 1,549 records · Page 86
All sourcesNASA NTRSDOE OSTI
DOE OSTI2016

Flow Field and Pressure Measurements in a 61-Pin Wire-Wrapped Bundle

Texas A and M University is conducting isothermal flow tests in a wire-wrapped, 61-pin hexagonal fuel bundle to support the research on advanced nuclear fuel development sponsored by the Department of Energy (DOE). The purpose of these tests is to produce high spatial- and temporal-resolution measurements of the flow and pressure measurements at different locations of the test bundle. The data resulting from these tests will be used to benchmark numerical simulations of the experiments using both industry-standard Reynolds-averaged Navier-Stokes (RANS) based Computation Fluid Dynamics (CFD) methods and a highly resolved Large Eddy Simulation (LES) computer code. This article described the main features of the experimental facility, its capabilities in producing experimental data of pressure drop and velocity fields at different axial locations and within different exterior and interior sub-channels, using advanced flow visualization techniques. Examples of experimental data outputs and typical locations of the measurements that will included in the research scope are also shown in this article. (authors)

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
DOE OSTI2016

Spacer Grid and Mixing Vanes Effects Study on Two-phase Flow Using Bubble Tracking Simulations

In Light Water Reactor (LWR) cores, the nuclear fuel rods are arranged together as rod bundles in a triangular or square pattern. The fuel rod bundles are supported by the spacer grids with mixing vanes. An understanding of spacer grid and mixing vanes effect is important since it has a direct impact on the turbulence structure in the flow and the heat removal efficiency from the fuel rod surface. Although a considerable amount of literature has been published on the experimental investigation of spacer grid and mixing vanes effect, the measurements were mainly focused on single-phase flow. The influence of spacer grid and mixing vanes on two-phase flow is yet to be fully understood. Some preliminary research efforts have been invested in nuclear community. For example, Wheeler et al. experimentally studied the effects of spacer grid in air-water two-phase flow. Although mixing vanes are not considered, they presented interesting results of time-averaged two-phase flow quantities at upstream and downstream locations of the spacer grid. Over the past decades, the rapid development of High-Performance Computing (HPC) is allowing for the increasingly larger-scale high-fidelity simulations. Thanks to this favorable historical trend, Direct Numerical Simulation (DNS), integrated with Interface Tracking Methods (ITM), has been emerging as a valuable tool to complement and expand our ability to understand the two-phase flow phenomenon. Meanwhile, in order to make best use of the large-scale two-phase flow simulations, advanced data analysis techniques are being developed, as they enable the extraction of detailed parameters about individual bubble behavior. The level set ITM is utilized in our DNS flow solver, which makes use of a signed distance field. The gas-liquid interface is modeled by the zero level set. The level set method can be readily used to distinguish different phases based on the sign of corresponding level set value (e.g. the sign is positive in the liquid phase while negative in the gas phase.). However, the traditional level set approach is not able to collect calculated values and associate with specific bubbles when multiple (even thousands of) bubbles present in the simulations of two-phase flow. This drawback hinders the collection of useful bubble information, which can give us valuable insights about bubbly flows. For example, how the different local fluid conditions could affect bubble interfacial forces, bubble deformation level, and eventually the bubble distribution throughout the whole domain. A bubble tracking methodology has been developed to address the drawbacks and improve the data collection efficiency of level set based two-phase flow simulations. In the present work, a single Pressurized Water Reactor (PWR) subchannel with reduced size spacer grid and mixing vanes is chosen as the computational domain. The springs and dimples of the spacer grid are not represented for simplicity in the present investigation. The bubble tracking methodology is applied to the simulations of two-phase bubbly flow through spacer grid and mixing vanes region. In the intended simulation, the liquid turbulence will be fully resolved by DNS while the two-phase behaviors are captured by level set ITM. Bubble tracking data are to be processed to investigate the spacer grid and mixing vanes effect on two-phase bubbly flow. (authors)

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗
DOE OSTI2012

Energy Innovation Hubs: A Home for Scientific Collaboration

Secretary Chu will host a live, streaming Q&A session with the directors of the Energy Innovation Hubs on Tuesday, March 6, at 2:15 p.m. EST. The directors will be available for questions regarding their teams' work and the future of American energy. Ask your questions in the comments below, or submit them on Facebook, Twitter (@energy), or send an e-mail to newmedia@hq.doe.gov, prior or during the live event. Dr. Hank Foley is the director of the Greater Philadelphia Innovation Cluster for Energy-Efficient Buildings, which is pioneering new data intensive techniques for designing and operating energy efficient buildings, including advanced computer modeling. Dr. Douglas Kothe is the director of the Consortium for Advanced Simulation of Light Water Reactors, which uses powerful supercomputers to create "virtual" reactors that will help improve the safety and performance of both existing and new nuclear reactors. Dr. Nathan Lewis is the director of the Joint Center for Artificial Photosynthesis, which focuses on how to produce fuels from sunlight, water, and carbon dioxide. The Energy Innovation Hubs are major integrated research centers, with researchers from many different institutions and technical backgrounds. Each hub is focused on a specific high priority goal, rapidly accelerating scientific discoveries and shortening the path from laboratory innovation to technological development and commercial deployment of critical energy technologies. Ask your questions in the comments below, or submit them on Facebook, Twitter (@energy), or send an e-mail to newmedia@energy.gov, prior or during the live event. The Energy Innovation Hubs are major integrated research centers, with researchers from many different institutions and technical backgrounds. Each Hub is focused on a specific high priority goal, rapidly accelerating scientific discoveries and shortening the path from laboratory innovation to technological development and commercial deployment of critical energy technologies. Dr. Hank Holey is the director of the Greater Philadelphia Innovation Cluster for Energy-Efficient Buildings, which is pioneering new data intensive techniques for designing and operating energy efficient buildings, including advanced computer modeling. Dr. Douglas Kothe is the director of the Modeling and Simulation for Nuclear Reactors Hub, which uses powerful supercomputers to create "virtual" reactors that will help improve the safety and performance of both existing and new nuclear reactors. Dr. Nathan Lewis is the director of the Joint Center for Artificial Photosynthesis Hub, which focuses on how to produce biofuels from sunlight, water, and carbon dioxide.

13 HYDRO ENERGY↗
DOE OSTI2023

On the potentially transformative role of auxiliary-field quantum Monte Carlo in quantum chemistry: A highly accurate method for transition metals and beyond

Approximate solutions to the ab initio electronic structure problem have been a focus of theoretical and computational chemistry research for much of the past century, with the goal of predicting relevant energy differences to within “chemical accuracy” (1 kcal/mol). For small organic molecules, or in general, for weakly correlated main group chemistry, a hierarchy of single-reference wave function methods has been rigorously established, spanning perturbation theory and the coupled cluster (CC) formalism. For these systems, CC with singles, doubles, and perturbative triples is known to achieve chemical accuracy, albeit at O(N7) computational cost. In addition, a hierarchy of density functional approximations of increasing formal sophistication, known as Jacob’s ladder, has been shown to systematically reduce average errors over large datasets representing weakly correlated chemistry. However, the accuracy of such computational models is less clear in the increasingly important frontiers of chemical space including transition metals and f-block compounds, in which strong correlation can play an important role in reactivity. A stochastic method, phaseless auxiliary-field quantum Monte Carlo (ph-AFQMC), has been shown to be capable of producing chemically accurate predictions even for challenging molecular systems beyond the main group, with relatively low O(N3 − N4) cost and near-perfect parallel efficiency. Herein, we present our perspectives on the past, present, and future of the ph-AFQMC method. We focus on its potential in transition metal quantum chemistry to be a highly accurate, systematically improvable method that can reliably probe strongly correlated systems in biology and chemical catalysis and provide reference thermochemical values (for future development of density functionals or interatomic potentials) when experiments are either noisy or absent. Finally, we discuss the present limitations of the method and where we expect near-term development to be most fruitful.

Chemistry↗
DOE OSTI2021

Big PanDa Workflow Management on Titan for High Energy and Nuclear Physics and for Future Extreme Scale Scientific Application

Over a three year period, from 2016-2019, this project demonstrated the scientific benefits of integrating the Titan supercomputer at Oak Ridge Leadership Computing Facility into traditional high throughput grid based distributed computing systems managed by PanDA, the workflow management system used for the execution of all distributed computing applications by the ATLAS experiment at the Large Hadron Collider. PanDA manages millions of batch jobs daily at hundreds of clusters worldwide on request by thousands of physicist users, and processes more than an exabyte of data annually using grid middleware. High levels of operational use of Titan was sustained by PanDA in order to meet the physics goals of ATLAS. The success of this project led to the use of other supercomputers worldwide by ATLAS, and to the adoption of PanDA by other experiments and other scientists. Multiple innovative operational and computer science research goals were achieved supporting the use of supercomputers for scientific domains with large scale distributed data and distributed processing needs.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

From ionic clusters dynamics to network constraints in ionic polymer solutions

Physical networks formed by ionizable polymers with ionic clusters as crosslinks are controlled by coupled dynamics that transcend from ionic clusters through chain motion to macroscopic response. Here, the coupled dynamics, across length scales, from the ionic clusters to the networks in toluene swollen polystyrene sulfonate networks, were directly correlated, as the electrostatic environment of the physical crosslinks was altered. The multiscale insight is attained by coupling neutron spin echo measurements with molecular dynamics simulations, carried out to times typical of relaxation of polymers in solutions. The experimental dynamic structure factor is in outstanding agreement with the one calculated from computer simulations, as the networks are perturbed by elevating the temperature and changing the electrostatic environment. In toluene, the long-lived clusters remain stable over hundreds of ns across a broad temperature range, while the polymer network remains dynamic. In conclusion, though the size of the clusters changes as the dielectric constant of the solvent is modified through the addition of ethanol, they remain stable but morph, enhancing the polymer chain dynamics.

36 MATERIALS SCIENCE↗
DOE OSTI2013

Rh-Based Mixed Alcohol Synthesis Catalysts: Characterization and Computational Report

The U.S. Department of Energy is conducting a program focused on developing a process for the conversion of biomass to bio-based fuels and co-products. Biomass-derived syngas is converted thermochemically within a temperature range of 240 to 330°C and at elevated pressure (e.g., 1200 psig) over a catalyst. Ethanol is the desired reaction product, although other side compounds are produced, including C 3 to C 5 alcohols; higher (i.e., greater than C 1 ) oxygenates such as methyl acetate, ethyl acetate, acetic acid and acetaldehyde; and higher hydrocarbon gases such as methane, ethane/ethene, propane/propene, etc. Saturated hydrocarbon gases (especially methane) are undesirable because they represent a diminished yield of carbon to the desired ethanol product and represent compounds that must be steam reformed at high energy cost to reproduce CO and H 2 . Ethanol produced by the thermochemical reaction of syngas could be separated and blended directly with gasoline to produce a liquid transportation fuel. Additionally, higher oxygenates and unsaturated hydrocarbon side products such as olefins also could be further processed to liquid fuels. The goal of the current project is the development of a Rh-based catalyst with high activity and selectivity to C 2 + oxygenates. This report chronicles an effort to characterize numerous supports and catalysts to identify particular traits that could be correlated with the most active and/or selective catalysts. Carbon and silica supports and catalysts were analyzed. Generally, analyses provided guidance in the selection of acceptable catalyst supports. For example, supports with high surface areas due to a high number of micropores were generally found to be poor at producing oxygenates, possibly because of mass transfer limitations of the products formed out of the micropores. To probe fundamental aspects of the complicated reaction network of CO with H 2 , a computational/ theoretical investigation using quantum mechanical and ab initio molecular dynamics calculations was initiated in 2009. Computational investigations were performed first to elucidate understanding of the nature of the catalytically active site. Thermodynamic calculations revealed that Mn likely exists as a metallic alloy with Rh in Rh-rich environments under reducing conditions at the temperatures of interest. After determining that reduced Rh-Mn alloy metal clusters were in a reduced state, the activation energy barriers of numerous transition state species on the catalytically active metal particles were calculated to compute the activation barriers of several reaction pathways that are possible on the catalyst surface. Comparison of calculations with a Rh nanoparticle versus a Rh-Mn nanoparticle revealed that the presence of Mn enabled the reaction pathway of CH with CO to form an adsorbed CHCO species, which was a precursor to C 2 + oxygenates. The presence of Mn did not have a significant effect on the rate of CH 4 production. Ir was observed during empirical catalyst screening experiments to improve the activity and selectivity of Rh-Mn catalysts. Thus, the addition of Ir to the Rh-Mn nanoparticles also was probed computationally. Simulations of Rh-Mn-Ir nanoparticles revealed that, with sufficient Ir concentrations, the Rh, Mn and Ir presumably would be well mixed within a nanoparticle. Activation barriers were calculated for Rh-Mn-Ir nanoparticles for several C-, H-, and O-containing transitional species on the nanoparticle surface. It was found that the presence of Ir opened yet another reactive pathway whereby HCO is formed and may undergo insertion with CHx surface moieties. The reaction pathway opened by the presence of Ir is in addition to the CO + CH pathway opened by the presence of Mn. Similar to Mn, the presence of Ir was not found to not affect the rate of CH 4 production.

catalyst↗
DOE OSTI2020

Distributed Computing for the Project 8 Experiment

The Project 8 collaboration aims to measure the absolute neutrino mass or improve on the current limit by measuring the tritium beta decay electron spectrum. We present the current distributed computing model for the Project 8 experiment. Project 8 is in its second phase of data taking with a near continuous data rate of 1Gbps. The current computing model uses DIRAC (Distributed Infrastructure with Remote Agent Control) for its workflow and data management. A detailed meta-data assignment using the DIRAC File Catalog is used to automate raw data transfers and subsequent stages of data processing. The DIRAC system is deployed on containers managed using a Kubernetes cluster to provide a scalable infrastructure. A modified DIRAC Site Director provides the ability to submit jobs using Singularity on opportunistic High-Performance Computing (HPC) sites.

Distributed Computing, Kubernetes, DIRAC, Project ↗
DOE OSTI2026

Geometry, spin coupling, and dielectric control of redox potentials in [4Fe–4S] Clusters

Iron–sulfur (Fe–S) clusters are common biological cofactors that facilitate vital redox reactions. Despite extensive research, the molecular basis of redox potential tuning in ferredoxin-like proteins remains an active area of debate. In this study, we combine statistical analysis of over one thousand [4Fe–4S]-containing protein structures from the Protein Data Bank (PDB) with broken-symmetry and extended broken-symmetry density functional theory to examine how cysteine ligand orientations and environmental screening affect redox properties of the clusters. We identified five main ligand configurations, three of which are predominant in natural structures. Among these, the adiabatic electron affinity differs by less than 0.1 V, indicating that, while geometry plays a secondary role, it allows localized fine-tuning of redox properties. In contrast, electrostatic and solvation effects primarily determine the overall potential range.

Computational Chemistry↗
DOE OSTI2012

Environment-based pin-power reconstruction method for homogeneous core calculations

Core calculation schemes are usually based on a classical two-step approach associated with assembly and core calculations. During the first step, infinite lattice assemblies calculations relying on a fundamental mode approach are used to generate cross-sections libraries for PWRs core calculations. This fundamental mode hypothesis may be questioned when dealing with loading patterns involving several types of assemblies (UOX, MOX), burnable poisons, control rods and burn-up gradients. This paper proposes a calculation method able to take into account the heterogeneous environment of the assemblies when using homogeneous core calculations and an appropriate pin-power reconstruction. This methodology is applied to MOX assemblies, computed within an environment of UOX assemblies. The new environment-based pin-power reconstruction is then used on various clusters of 3x3 assemblies showing burn-up gradients and UOX/MOX interfaces, and compared to reference calculations performed with APOLLO-2. The results show that UOX/MOX interfaces are much better calculated with the environment-based calculation scheme when compared to the usual pin-power reconstruction method. The power peak is always better located and calculated with the environment-based pin-power reconstruction method on every cluster configuration studied. This study shows that taking into account the environment in transport calculations can significantly improve the pin-power reconstruction so far as it is consistent with the core loading pattern. (authors)

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2018

Low-lying excited states by constrained DFT

Exploiting the machinery of Constrained Density Functional Theory (CDFT), we propose a variational method for calculating low-lying excited states of molecular systems. We dub this method eXcited CDFT (XCDFT). Excited states are obtained by self-consistently constraining a user-defined population of electrons, N c , in the virtual space of a reference set of occupied orbitals. By imposing this population to be N c = 1.0, we computed the first excited state of 15 molecules from a test set. Our results show that XCDFT achieves an accuracy in the predicted excitation energy only slightly worse than linear-response time-dependent DFT (TDDFT), but without incurring into problems of variational collapse typical of the more commonly adopted ΔSCF method. In addition, we selected a few challenging processes to test the limits of applicability of XCDFT. We find that in contrast to TDDFT, XCDFT is capable of reproducing energy surfaces featuring conical intersections (azobenzene and H3) with correct topology and correct overall energetics also away from the intersection. Venturing to condensed-phase systems, XCDFT reproduces the TDDFT solvatochromic shift of benzaldehyde when it is embedded by a cluster of water molecules. Finally, we find XCDFT to be a competitive method among single-reference methods for computations of excited states in terms of time to solution, rate of convergence, and accuracy of the result.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2017

Imputation for multisource data with comparison and assessment techniques

Missing data are prevalent issue in analyses involving data collection. The problem of missing data is exacerbated for multisource analysis, where data from multiple sensors are combined to arrive at a single conclusion. In this scenario, it is more likely to occur and can lead to discarding a large amount of data collected; however, the information from observed sensors can be leveraged to estimate those values not observed. We propose two methods for imputation of multisource data, both of which take advantage of potential correlation between data from different sensors, through ridge regression and a state-space model. These methods, as well as the common median imputation, are applied to data collected from a variety of sensors monitoring an experimental facility. Performance of imputation methods is compared with the mean absolute deviation; however, rather than using this metric to solely rank themethods,we also propose an approach to identify significant differences. Imputation techniqueswill also be assessed by their ability to produce appropriate confidence intervals, through coverage and length, around the imputed values. Finally, performance of imputed datasets is compared with a marginalized dataset through a weighted k-means clustering. In general, we found that imputation through a dynamic linearmodel tended to be the most accurate and to produce the most precise confidence intervals, and that imputing the missing values and down weighting them with respect to observed values in the analysis led to the most accurate performance.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2014

GoFFish: A Sub-Graph Centric Framework for Large-Scale Graph Analytics1

Large scale graph processing is a major research area for Big Data exploration. Vertex centric programming models like Pregel are gaining traction due to their simple abstraction that allows for scalable execution on distributed systems naturally. However, there are limitations to this approach which cause vertex centric algorithms to under-perform due to poor compute to communication overhead ratio and slow convergence of iterative superstep. In this paper we introduce GoFFish a scalable sub-graph centric framework co-designed with a distributed persistent graph storage for large scale graph analytics on commodity clusters. We introduce a sub-graph centric programming abstraction that combines the scalability of a vertex centric approach with the flexibility of shared memory sub-graph computation. We map Connected Components, SSSP and PageRank algorithms to this model to illustrate its flexibility. Further, we empirically analyze GoFFish using several real world graphs and demonstrate its significant performance improvement, orders of magnitude in some cases, compared to Apache Giraph, the leading open source vertex centric implementation. We map Connected Components, SSSP and PageRank algorithms to this model to illustrate its flexibility. Further, we empirically analyze GoFFish using several real world graphs and demonstrate its significant performance improvement, orders of magnitude in some cases, compared to Apache Giraph, the leading open source vertex centric implementation.

Simmhan, Yogesh↗
DOE OSTI2015

Computational Study of Ethanol Conversion on Al 8 O 12 as a Model for γ-Al 2 O 3

Correlated molecular orbital theory at the coupled cluster CCSD(T) level with density functional theory geometries is used to study ethanol dehydration, dehydrogenation, and condensation reactions on an the Al8O12 cluster which is a model for γ-Al2O3. The Al in the active site on the cluster is a strong Lewis acid. The reactions begin with formation of a very stable Lewis acid–base ethanol–cluster adduct. Dehydration proceeds by β-H transfer to a bicoordinate oxygen leading to the direct formation of ethylene and two OH groups following an E2 mechanism. Dehydrogenation proceeds directly by α-H transfer to the active metal center and a proton transfer to a bicoordinate bridge O to form acetaldehyde plus a metal hydride and a hydroxyl, again an E2 mechanism. After addition of a second ethanol, diethyl ether is generated by an α-C transfer from the first to the second ethanol, an acid-driven SN2 mechanism. Condensation and dehydration with two alcohols have comparable energy barriers. The addition of a second ethanol or a water molecule raises the energy barriers. Condensation and dehydration are predicted to be more likely than dehydrogenation. The computational results for the mechanism and the energetics agree well with the available experimental data.

Fang, Zongtang↗
DOE OSTI2024

Clustering and Cliques in Preferential Attachment Random Graphs with Edge Insertion

In this paper, we investigate the global clustering coefficient (a.k.a transitivity) and clique number of graphs generated by a preferential attachment random graph model with an additional feature of allowing edge connections between existing vertices. Specifically, at each time step t, either a new vertex is added with probability f(t), or an edge is added between two existing vertices with probability 1 – f(t). We establish concentration inequalities for the global clustering and clique number of the resulting graphs under the assumption that f(t) is a regularly varying function at infinity with index of regular variation –$\gamma$, where $\gamma$ $\in$ [0, 1). Finally, we also demonstrate an inverse relation between these two statistics: the clique number is essentially the reciprocal of the global clustering coefficient.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2015

Big Data Analytics for Demand Response: Clustering Over Space and Time

The pervasive deployment of advanced sensing infrastructure in Cyber-Physical systems, such as the Smart Grid, has resulted in an unprecedented data explosion. Such data exhibit both large volumes and high velocity characteristics, two of the three pillars of Big Data, and have a time-series notion as datasets in this context typically consist of successive measurements made over a time interval. Time-series data can be valuable for data mining and analytics tasks such as identifying the “right” customers among a diverse population, to target for Demand Response programs. However, time series are challenging to mine due to their high dimensionality. In this paper, we motivate this problem using a real application from the smart grid domain. We explore novel representations of time-series data for BigData analytics, and propose a clustering technique for determining natural segmentation of customers and identification of temporal consumption patterns. Our method is generizable to large-scale, real-world scenarios, without making any assumptions about the data. We evaluate our technique using real datasets from smart meters, totaling ~ 18,200,000 data points, and show the efficacy of our technique in efficiency detecting the number of optimal number of clusters.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2019

A space for lattice representation and clustering

Algorithms for quantifying the differences between two lattices are used for Bravais lattice determination, database lookup for unit cells to select candidates for molecular replacement, and recently for clustering to group together images from serial crystallography. It is particularly desirable for the differences between lattices to be computed as a perturbation-stable metric, i.e. as distances that satisfy the triangle inequality, so that standard tree-based nearest-neighbor algorithms can be used, and for which small changes in the lattices involved produce small changes in the distances computed. A perturbation-stable metric space related to the reduction algorithm of Selling and to the Bravais lattice determination methods of Delone is described. Two ways of representing the space, as six-dimensional real vectors or equivalently as three-dimensional complex vectors, are presented and applications of these metrics are discussed. (Note: in his later publications, Boris Delaunay used the Russian version of his surname, Delone.)

36 MATERIALS SCIENCE↗
DOE OSTI2020

Optimizing High Performance Markov Clustering for Pre-Exascale Architectures

HipMCL is a high-performance distributed memory implementation of the popular Markov Cluster Algorithm (MCL) and can cluster large-scale networks within hours using a few thousand CPU-equipped nodes. It relies on sparse matrix computations and heavily makes use of the sparse matrix-sparse matrix multiplication kernel (SpGEMM). The existing parallel algorithms in HipMCL are not scalable to Exascale architectures, both due to their communication costs dominating the runtime at large concurrencies and also due to their inability to take advantage of accelerators that are increasingly popular. In this work, we systematically remove scalability and performance bottlenecks of HipMCL. We enable GPUs by performing the expensive expansion phase of the MCL algorithm on GPU. Additionally, we propose a CPU-GPU joint distributed SpGEMM algorithm called pipelined Sparse SUMMA and integrate a probabilistic memory requirement estimator that is fast and accurate. Furthermore, we develop a new merging algorithm for the incremental processing of partial results produced by the GPUs, which improves the overlap efficiency and the peak memory usage. We also integrate a recent and faster algorithm for performing SpGEMM on CPUs. We validate our new algorithms and optimizations with extensive evaluations. With the enabling of the GPUs and integration of new algorithms, HipMCL is up to 12.4x faster, being able to cluster a network with 70 million proteins and 68 billion connections just under 15 minutes using 1024 nodes of ORNL's Summit supercomputer.

97 MATHEMATICS AND COMPUTING↗