Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Distributed Memory Parallel”

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 721 records · Page 40

Kanerva's sparse distributed memory with multiple hamming thresholds

If the stored input patterns of Kanerva's Sparse Distributed Memory (SDM) are highly correlated, utilization of the storage capacity is very low compared to the case of uniformly distributed random input patterns. We consider a variation of SDM that has a better storage capacity utilization for correlated input patterns. This approach uses a separate selection threshold for each physical storage address or hard location. The selection of the hard locations for reading or writing can be done in parallel of which SDM implementations can benefit.

Pohja, Seppo↗

Distributed Cross-Channel Hierarchical Aggregation for Foundation Models

Vision-based scientific foundation models hold significant promise for advancing scientific discovery and innovation. This potential stems from their ability to aggregate images from diverse sources—such as varying physical groundings or data acquisition systems—and to learn spatio-temporal correlations using transformer architectures. However, tokenizing and aggregating images can be compute-intensive, a challenge not fully addressed by current distributed methods. In this work, we introduce the Distributed Cross-Channel Hierarchical Aggregation (D-CHAG) approach designed for datasets with a large number of channels across image modalities. Our method is compatible with any model-parallel strategy and any type of vision transformer architecture, significantly improving computational efficiency. We evaluated D-CHAG on hyperspectral imaging and weather forecasting tasks. When integrated with tensor parallelism and model sharding, our approach achieved up to a 75% reduction in memory usage and more than doubled sustained throughput on up to 1,024 AMD GPUs on the Frontier Supercomputer.

Tsaris, Aristeidis (aris) [ORNL] (ORCID:0000000277↗

Planning for Pre-Exascale Platform Environment (Fiscal Year 2015 Level 2 Milestone 5216)

This Plan for ASC Pre-Exascale Platform Environments document constitutes the deliverable for the fiscal year 2015 (FY15) Advanced Simulation and Computing (ASC) Program Level 2 milestone Planning for Pre-Exascale Platform Environment. It acknowledges and quantifies challenges and recognized gaps for moving the ASC Program towards effective use of exascale platforms and recommends strategies to address these gaps. This document also presents an update to the concerns, strategies, and plans presented in the FY08 predecessor document that dealt with the upcoming (at the time) petascale high performance computing (HPC) platforms. With the looming push towards exascale systems, a review of the earlier document was appropriate in light of the myriad architectural choices currently under consideration. The ASC Program believes the platforms to be fielded in the 2020s will be fundamentally different systems that stress ASC’s ability to modify codes to take full advantage of new or unique features. In addition, the scale of components will increase the difficulty of maintaining an errorfree system, thus driving new approaches to resilience and error detection/correction. The code revamps of the past, from serial- to vector-centric code to distributed memory to threaded implementations, will be revisited as codes adapt to a new message passing interface (MPI) plus “x” or more advanced and dynamic programming models based on architectural specifics. Development efforts are already underway in some cases, and more difficult or uncertain aspects of the new architectures will require research and analysis that may inform future directions for program choices. In addition, the potential diversity of system architectures may require parallel if not duplicative efforts to analyze and modify environments, codes, subsystems, libraries, debugging tools, and performance analysis techniques as well as exploring new monitoring methodologies. It is difficult if not impossible to selectively eliminate some of these activities until more information is available through simulations of potential architectures, analysis of systems designs, and informed study of commodity technologies that will be the constituent parts of future platforms.

97 MATHEMATICS AND COMPUTING↗

Computing the Properties of Matter with Leadership Computing Resources (Closeout Report for DE-SC0018121)

In order to add more capabilities to Halide, we have designed a new framework called Tiramisu and integrated this framework into Halide. Since Tiramisu enables Halide to target heterogeneous architectures, our development efforts have been refocused on Tiramisu. Most high-performance computer systems today are complex and increasingly heterogeneous; they may have CPUs, GPUs and FPGAs. Achieving best performance requires taking full advantage of all these different architectures. To address this issue, we have designed Tiramisu, an optimization framework that enables Halide (and other DSLs) to target heterogeneous architectures. Tiramisu is an optimization framework that takes as input a high level, architecture-independent representation of code and a set of scheduling and data mapping commands that guide code transformation. The input can either be generated by a domain-specific language (DSL) compiler such as Halide or directly written by a programmer. Tiramisu then applies the user-specified code and data-layout transformations and generates an architecture-specific, low-level intermediate representation (IR) that takes advantage of modern architectural features such as multicore parallelism, non-uniform memory (NUMA) hierarchies, clusters, and accelerators like GPUs and FPGAs. We integrated Tiramisu within Halide and implemented a representative set of benchmarks to evaluate this integration. Tiramisu is now open source and is available for public use (http://tiramisu-compiler.org/). A paper about Tiramisu was published, it shows that Tiramisu extends Halide with many new capabilities and that Tiramisu can generate efficient code for multicores, GPUs, FPGAs and distributed heterogeneous systems. The performance of code generated by the Tiramisu backends matches or exceeds hand optimized reference implementations. For example, the multicore backend matches the highly optimized Intel MKL library on many kernels and shows speedups reaching 4x over the original Halide. In addition to making Tiramisu more robust, we have used Tiramisu to implement a set of representative tensor operation for constructing baryon building blocks required for multi baryon contractions in LQCD. In order to implement this code, we needed to generalize Tiramisu in two ways: first we needed to support indirect array accesses, and second, we needed to add support for complex numbers to Tiramisu. The code generated by Tiramisu is 6x faster than the reference code. Our efforts towards an MPI based multi-node version of tiramisu have matured and the resulting code scales well on multiple nodes (tests up to 512 KNL nodes have been undertaken).

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

An Efficient Variable-Length Data-Compression Scheme

Adaptive variable-length coding scheme for compression of stream of independent and identically distributed source data involves either Huffman code or alternating run-length Huffman (ARH) code, depending on characteristics of data. Enables efficient compression of output of lossless or lossy precompression process, with speed and simplicity greater than those of older coding schemes developed for same purpose. In addition, scheme suitable for parallel implementation on hardware with modular structure, provides for rapid adaptation to changing data source, compatible with block orientation to alleviate memory requirements, ensures efficiency over wide range of entropy, and easily combined with such other communication schemes as those for containment of errors and for packetization.

Cheung, Kar-Ming↗

Load Balancing Unstructured Adaptive Grids for CFD Problems

Mesh adaption is a powerful tool for efficient unstructured-grid computations but causes load imbalance among processors on a parallel machine. A dynamic load balancing method is presented that balances the workload across all processors with a global view. After each parallel tetrahedral mesh adaption, the method first determines if the new mesh is sufficiently unbalanced to warrant a repartitioning. If so, the adapted mesh is repartitioned, with new partitions assigned to processors so that the redistribution cost is minimized. The new partitions are accepted only if the remapping cost is compensated by the improved load balance. Results indicate that this strategy is effective for large-scale scientific computations on distributed-memory multiprocessors.

Biswas, Rupak↗

Distributed parallel messaging for multiprocessor systems

A method and apparatus for distributed parallel messaging in a parallel computing system. The apparatus includes, at each node of a multiprocessor network, multiple injection messaging engine units and reception messaging engine units, each implementing a DMA engine and each supporting both multiple packet injection into and multiple reception from a network, in parallel. The reception side of the messaging unit (MU) includes a switch interface enabling writing of data of a packet received from the network to the memory system. The transmission side of the messaging unit, includes switch interface for reading from the memory system when injecting packets into the network.

97 MATHEMATICS AND COMPUTING↗

A Step towards Energy Efficient Computing: Redesigning a Hydrodynamic Application on CPU-GPU, In: 2014 IEEE 28th International Parallel and Distributed Processing Symposium

Power and energy consumption are becoming an increasing concern in high performance computing. Compared to multi-core CPUs, GPUs have a much better performance per watt. In this paper we discuss efforts to redesign the most computation intensive parts of BLAST, an application that solves the equations for compressible hydrodynamics with high order finite elements, using GPUs BLAST, Dobrev. In order to exploit the hardware parallelism of GPUs and achieve high performance, we implemented custom linear algebra kernels. We intensively optimized our CUDA kernels by exploiting the memory hierarchy, which exceed the vendor's library routines substantially in performance. We proposed an auto tuning technique to adapt our CUDA kernels to the orders of the finite element method. Compared to a previous base implementation, our redesign and optimization lowered the energy consumption of the GPU in two aspects: 60% less time to solution and 10% less power required. Compared to the CPU-only solution, our GPU accelerated BLAST obtained a 2.5× overall speedup and 1.42× energy efficiency (green up) using 4th order (Q_4) finite elements, and a 1.9× speedup and 1.27× green up using 2nd order (Q2) finite elements.

Computer Science↗

Parallel accelerated cyclic reduction preconditioner for three-dimensional elliptic PDEs with variable coefficients

We present a robust and scalable preconditioner for the solution of large-scale linear systems that arise from the discretization of elliptic PDEs amenable to rank compression. The preconditioner is based on hierarchical low-rank approximations and the cyclic reduction method. The setup and application phases of the preconditioner achieve log-linear complexity in memory footprint and number of operations, and numerical experiments exhibit good weak and strong scalability at large processor counts in a distributed memory environment. Numerical experiments with linear systems that feature symmetry and nonsymmetry, definiteness and indefiniteness, constant and variable coefficients demonstrate the preconditioner applicability and robustness. Furthermore, it is possible to control the number of iterations via the accuracy threshold of the hierarchical matrix approximations and their arithmetic operations, and the tuning of the admissibility condition parameter. Together, these parameters allow for optimization of the memory requirements and performance of the preconditioner.

97 MATHEMATICS AND COMPUTING↗

Performance Analysis and Optimal Node-aware Communication for Enlarged Conjugate Gradient Methods

Krylov methods are a key way of solving large sparse linear systems of equations but suffer from poor strong scalability on distributed memory machines. Furthermore, this is due to high synchronization costs from large numbers of collective communication calls alongside a low computational workload. Enlarged Krylov methods address this issue by decreasing the total iterations to convergence, an artifact of splitting the initial residual and resulting in operations on block vectors. In this article, we present a performance study of an enlarged Krylov method, Enlarged Conjugate Gradients (ECG), noting the impact of block vectors on parallel performance at scale. Most notably, we observe the increased overhead of point-to-point communication as a result of denser messages in the sparse matrix-block vector multiplication kernel. Additionally, we present models to analyze expected performance of ECG, as well as motivate design decisions. Most importantly, we introduce a new point-to-point communication approach based on node-aware communication techniques that increases efficiency of the method at scale.

97 MATHEMATICS AND COMPUTING↗

A Multiple Sphere T-Matrix Fortran Code for Use on Parallel Computer Clusters

A general-purpose Fortran-90 code for calculation of the electromagnetic scattering and absorption properties of multiple sphere clusters is described. The code can calculate the efficiency factors and scattering matrix elements of the cluster for either fixed or random orientation with respect to the incident beam and for plane wave or localized- approximation Gaussian incident fields. In addition, the code can calculate maps of the electric field both interior and exterior to the spheres.The code is written with message passing interface instructions to enable the use on distributed memory compute clusters, and for such platforms the code can make feasible the calculation of absorption, scattering, and general EM characteristics of systems containing several thousand spheres.

Mackowski, D. W.↗

Building Scalable PGAS Communication Subsystem on Blue Gene/Q

This paper presents a design of scalable Partitioned Global Address Space (PGAS) communication subsystems on recently proposed Blue Gene/Q architecture. The proposed design provides an in-depth modeling of communication infrastructure using Parallel Active Messaging Interface (PAMI). The communication infrastructure is used to design time-space efficient communication protocols for frequently used data-types (contiguous, uniformly non-contiguous) using Remote Direct Memory Access (RDMA) get/put primitives. The proposed design accelerates load balance counters by using asynchronous threads, which are required due to the missing network hardware support for Atomic Memory Operations (AMOs). Under the proposed design, the synchronization traffic is reduced by tracking conflicting memory accesses in distributed space with slightly increment in space complexity. An evaluation with simple communication benchmarks show a adjacent node get latency of 2.89$\mu$s and peak bandwidth of 1775 MB/s resulting in $\approx$ 99\% communication efficiency.The evaluation shows a reduction in the execution time by up to 30\% for NWChem self consistent field calculation on 4096 processes using the proposed asynchronous thread based design.

Vishnu, Abhinav↗

Variably Saturated Flow and Multicomponent Biogeochemical Reactive Transport Modeling of a Uranium Bioremediation Field Experiment

Field experiments at a former uranium mill tailings site have identified the potential for stimulating indigenous bacteria to catalyze the conversion of aqueous uranium in the +6 oxidation state to immobile solid-associated uranium in the +4 oxidation state. This effectively removes uranium from solution resulting in groundwater concentrations below actionable standards. Three-dimensional, coupled variably-saturated flow and biogeochemical reactive transport modeling of a 2008 in situ uranium bioremediation field experiment is used to better understand the interplay of transport rates and biogeochemical reaction rates that determine the location and magnitude of key reaction products. A comprehensive reaction network, developed largely through previous 1-D modeling studies, was used to simulate the impacts on uranium behavior of pulsed acetate amendment, seasonal water table variation, spatially-variable physical (hydraulic conductivity, porosity) and geochemical (reactive surface area) material properties. A principal challenge is the mechanistic representation of biologically-mediated terminal electron acceptor process (TEAP) reactions whose products significantly alter geochemical controls on uranium mobility through increases in pH, alkalinity, exchangeable cations, and highly reactive reduction products. In general, these simulations of the 2008 Big Rusty acetate biostimulation field experiment in Rifle, Colorado confirmed previously identified behaviors including (1) initial dominance by iron reducing bacteria that concomitantly reduce aqueous U(VI), (2) sulfate reducing bacteria that become dominant after {approx}30 days and outcompete iron reducers for the acetate electron donor, (3) continuing iron-reducer activity and U(VI) bioreduction during dominantly sulfate reducing conditions, and (4) lower apparent U(VI) removal from groundwater during dominantly sulfate reducing conditions. New knowledge on simultaneously active metal and sulfate reducers has been incorporated into the modeling. In this case, an initially small population of slow growing sulfate reducers is active from the initiation of biostimulation. Three-dimensional, variably saturated flow modeling was used to address impacts of a falling water table during acetate injection. These impacts included a significant reduction in aquifer saturated thickness and isolation of residual reactants and products, as well as unmitigated uranium, in the newly unsaturated vadose zone. High permeability sandy gravel structures resulted in locally high flow rates in the vicinity of injection wells that increased acetate dilution. In downgradient locations, these structures created preferential flow paths for acetate delivery that enhanced local zones of TEAP reactivity and subsidiary reactions. Conversely, smaller transport rates associated with the lower permeability lithofacies (e.g., fine) and vadose zone were shown to limit acetate access and reaction. Once accessed by acetate, however, these same zones limited subsequent acetate dilution and provided longer residence times that resulted in higher concentrations of TEAP products when terminal electron donors and acceptors were not limiting. Finally, facies-based porosity and reactive surface area variations were shown to affect aqueous uranium concentration distributions; however, the ranges were sufficiently small to preserve general trends. Large computer memory and high computational performance were required to simulate the detailed coupled process models for multiple biogeochemical components in highly resolved heterogeneous materials for the 110-day field experiment and 50 days of post-biostimulation behavior. In this case, a highly-scalable subsurface simulator operating on 128 processor cores for 12 hours was used to simulate each realization. An equivalent simulation without parallel processing would have taken 60 days, assuming sufficient memory was available.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

Design and Implementation of Papyrus: Parallel Aggregate Persistent Storage

A surprising development in recently announced HPC platforms is the addition of, sometimes massive amounts of, persistent (nonvolatile) memory (NVM) in order to increase memory capacity and compensate for plateauing I/O capabilities. However, there are no portable and scalable programming interfaces using aggregate NVM effectively. This paper introduces Papyrus: a new software system built to exploit emerging capability of NVM in HPC architectures. Papyrus (or Parallel Aggregate Persistent -YRU- Storage) is a novel programming system that provides features for scalable, aggregate, persistent memory in an extreme-scale system for typical HPC usage scenarios. Papyrus mainly consists of Papyrus Virtual File System (VFS) and Papyrus Template Container Library (TCL). Papyrus VFS provides a uniform aggregate NVM storage image across diverse NVM architectures. It enables Papyrus TCL to provide a portable and scalable high-level container programming interface whose data elements are distributed across multiple NVM nodes without requiring the user to handle complex communication, synchronization, replication, and consistency model. We evaluate Papyrus on two HPC systems, including UTK Beacon and NERSC Cori, using real NVM storage devices.

Kim, Jungwon↗

Persistent Octrees for Parallel Mesh Refinement Through Non-Volatile Byte-Addressable Memory

We report that octree-based mesh adaptation has enabled simulations of complex physical phenomena. Existing meshing algorithms were proposed with the assumption that computer memory is volatile. Consequently, for failure recovery, the in-core algorithms need to save memory states as snapshots with slow file I/Os. The out-of-core algorithms store octants on disks for persistence. However, neither of them was designed to leverage unique characteristics of non-volatile byte-addressable memory (NVBM). In this paper, we propose a novel data structure Distributed Persistent Merged octree (DPM-octree) for both meshing and in-memory storage of persistent octrees using NVBM. It is a multi-version data structure and can recover from failures using its earlier persistent version stored in NVBM. In addition, we design a feature-directed sampling approach to help dynamically transform the DPM-octree layout for reducing NVBM-induced memory write latency. DPM-octree uses parity trees which are created using erasure coding and stored in NVBM to support low-latency in-memory octant recovery after data loss. DPM-octree has been successfully integrated with Gerris software for simulation of fluid dynamics. Finally, our experimental results with real-world scientific workloads show that DPM-octree scales up to 1.1 billion mesh elements with 1000 processors on the Titan supercomputer.

97 MATHEMATICS AND COMPUTING↗

A message passing kernel for the hypercluster parallel processing test bed

A Message-Passing Kernel (MPK) for the Hypercluster parallel-processing test bed is described. The Hypercluster is being developed at the NASA Lewis Research Center to support investigations of parallel algorithms and architectures for computational fluid and structural mechanics applications. The Hypercluster resembles the hypercube architecture except that each node consists of multiple processors communicating through shared memory. The MPK efficiently routes information through the Hypercluster, using a message-passing protocol when necessary and faster shared-memory communication whenever possible. The MPK also interfaces all of the processors with the Hypercluster operating system (HYCLOPS), which runs on a Front-End Processor (FEP). This approach distributes many of the I/O tasks to the Hypercluster processors and eliminates the need for a separate I/O support program on the FEP.

Blech, Richard A.↗

Scalable Work Stealing

Irregular and dynamic parallel applications pose significant challenges to achieving scalable performance on large-scale multicore clusters. These applications often require ongoing, dynamic load balancing in order to maintain efficiency. While effective at small scale, centralized load balancing schemes quickly become a bottleneck on large-scale clusters. Work stealing is a popular approach to distributed dynamic load balancing; however its performance on large-scale clusters is not well understood. Prior work on work stealing has largely focused on shared memory machines. In this work we investigate the design and scalability of work stealing on modern distributed memory systems. We demonstrate high efficiency and low overhead when scaling to 8,192 processors for three benchmark codes: a producer-consumer benchmark, the unbalanced tree search benchmark, and a multiresolution analysis kernel.

99 GENERAL AND MISCELLANEOUS↗

Sequence length scaling in vision transformers for scientific images on frontier

Vision Transformers (ViTs) are pivotal for foundational models in scientific imagery, including Earth science applications, due to their capability to process large sequence lengths. While transformers for text have inspired scaling sequence lengths in ViTs, adapting these for ViTs introduces unique challenges. We develop distributed sequence parallelism for ViTs, enabling them to handle up to 1M tokens. Our approach, leveraging DeepSpeed-Ulysses and Long-Sequence-Segmentation with model sharding, is the first to apply sequence parallelism in ViT training, achieving a 94% batch scaling efficiency on 2,048 AMD-MI250X GPUs. Evaluating sequence parallelism in ViTs, particularly in models up to 10B parameters, highlighted substantial bottlenecks. We countered these with hybrid sequence, pipeline, and flash attention strategies, to scale beyond single GPU memory limits. Our method significantly enhances climate modeling accuracy by 20% in temperature predictions, marking the first training of a vision transformer model to convergence with a sequence length of 188K tokens, using full self-attention.

Tsaris, Aristeidis (aris) [ORNL] (ORCID:0000000277↗