Search NASA⌕ Search

SEARCH · Search NASA

Results for “data 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 37 records · Page 2
All sourcesNASA NTRSDOE OSTI
NASA NTRS1994

Extending HPF for advanced data parallel applications

The stated goal of High Performance Fortran (HPF) was to 'address the problems of writing data parallel programs where the distribution of data affects performance'. After examining the current version of the language we are led to the conclusion that HPF has not fully achieved this goal. While the basic distribution functions offered by the language - regular block, cyclic, and block cyclic distributions - can support regular numerical algorithms, advanced applications such as particle-in-cell codes or unstructured mesh solvers cannot be expressed adequately. We believe that this is a major weakness of HPF, significantly reducing its chances of becoming accepted in the numeric community. The paper discusses the data distribution and alignment issues in detail, points out some flaws in the basic language, and outlines possible future paths of development. Furthermore, we briefly deal with the issue of task parallelism and its integration with the data parallel paradigm of HPF.

Chapman, Barbara↗
DOE OSTI2022

Accelerating Collective Communication in Data Parallel Training across Deep Learning Frameworks

This work develops new techniques within Horovod, a generic communication library supporting data parallel training across deep learning frameworks. In particular, we improve the Horovod control plane by implementing a new coordination scheme that takes advantage of the characteristics of the typical data parallel training paradigm, namely the repeated execution of collectives on the gradients of a fixed set of tensors. Using a caching strategy, we execute Horovod’s existing coordinator-worker logic only once during a typical training run, replacing it with a more efficient decentralized orchestration strategy using the cached data and a global intersection of a bitvector for the remaining training duration. Next, we introduce a feature for end users to explicitly group collective operations, enabling finer grained control over the communication buffer sizes. To evaluate our proposed strategies, we conduct experiments on a world-class supercomputer — Summit. We compare our proposals to Horovod’s original design and observe 2x performance improvement at a scale of 6000 GPUs; we also compare them against tf.distribute and torch.DDP and achieve 12% better and comparable performance, respectively, using up to 1536 GPUs; we compare our solution against BytePS in typical HPC settings and achieve about 20% better performance on a scale of 768 GPUs. Finally, we test our strategies on a scientific application (STEMDL) using up to 27,600 GPUs (the entire Summit) and show that we achieve a near-linear scaling of 0.93 with a sustained performance of 1.54 exaflops (with standard error +- 0.02) in FP16 precision.

Romero, Joshua↗
DOE OSTI2020

DeepClone: Lightweight State Replication of Deep Learning Models for Data Parallel Training

Training modern deep neural network (DNN) models involves complex workflows triggered by model exploration, sensitivity analysis, explainability, etc. A key primitive in this context is the ability to clone a model training instance, i.e. “fork” the training process in a potentially different direction, which enables comparisons of different evolution paths using variations of training data and model parameters. However, in a quest improve the training throughput, a mix of data parallel, model parallel, pipeline parallel and layer-wise parallel approaches are making the problem of cloning highly complex. In this paper, we explore the problem of efficient cloning under such circumstances. To this end, we leverage several properties of data-parallel training and layer-wise parallelism to design DeepClone, a cloning approach based on augmenting the execution graph to gain direct access to tensors, which are then sharded and reconstructed asynchronously in order to minimize runtime overhead, standby duration, readiness duration. Compared with state-of-art approaches, DeepClone shows orders of magnitude improvement for several classes of DNN models.

data-parallel training↗
DOE OSTI2019

Scalable Contour Tree Computation by Data Parallel Peak Pruning

As data sets grow to exascale, automated data analysis and visualization are increasingly important, to intermediate human understanding and to reduce demands on disk storage via in situ analysis. Trends in architecture of high performance computing systems necessitate analysis algorithms to make effective use of combinations of massively multicore and distributed systems. One of the principal analytic tools is the contour tree, which analyses relationships between contours to identify features of more than local importance. Unfortunately, the predominant algorithms for computing the contour tree are explicitly serial, and founded on serial metaphors, which has limited the scalability of this form of analysis. While there is some work on distributed contour tree computation, and separately on hybrid GPU-CPU computation, there is no efficient algorithm with strong formal guarantees on performance allied with fast practical performance. Furthermore, we report the first shared SMP algorithm for fully parallel contour tree computation, with formal guarantees of O(lg V lg t) parallel steps and O(V lg V) work for data with V samples and t contour tree supernodes, and implementations with more than 30× parallel speed up on both CPU using TBB and GPU using Thrust and up 70× speed up compared to the serial sweep and merge algorithm.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1991

Rectilinear partitioning of irregular data parallel computations

New mapping algorithms for domain oriented data-parallel computations, where the workload is distributed irregularly throughout the domain, but exhibits localized communication patterns are described. Researchers consider the problem of partitioning the domain for parallel processing in such a way that the workload on the most heavily loaded processor is minimized, subject to the constraint that the partition be perfectly rectilinear. Rectilinear partitions are useful on architectures that have a fast local mesh network. Discussed here is an improved algorithm for finding the optimal partitioning in one dimension, new algorithms for partitioning in two dimensions, and optimal partitioning in three dimensions. The application of these algorithms to real problems are discussed.

Nicol, David M.↗
DOE OSTI2017

Maximal clique enumeration with data-parallel primitives

The enumeration of all maximal cliques in an undirected graph is a fundamental problem arising in several research areas. We consider maximal clique enumeration on shared-memory, multi-core architectures and introduce an approach consisting entirely of data-parallel operations, in an effort to achieve efficient and portable performance across different architectures. We study the performance of the algorithm via experiments varying over benchmark graphs and architectures. Overall, we observe that our algorithm achieves up to a 33-time speedup and 9-time speedup over state-of-the-art distributed and serial algorithms, respectively, for graphs with higher ratios of maximal cliques to total cliques. Further, we attain additional speedups on a GPU architecture, demonstrating the portable performance of our data-parallel design.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1994

Array distribution in data-parallel programs

We consider distribution at compile time of the array data in a distributed-memory implementation of a data-parallel program written in a language like Fortran 90. We allow dynamic redistribution of data and define a heuristic algorithmic framework that chooses distribution parameters to minimize an estimate of program completion time. We represent the program as an alignment-distribution graph. We propose a divide-and-conquer algorithm for distribution that initially assigns a common distribution to each node of the graph and successively refines this assignment, taking computation, realignment, and redistribution costs into account. We explain how to estimate the effect of distribution on computation cost and how to choose a candidate set of distributions. We present the results of an implementation of our algorithms on several test problems.

Chatterjee, Siddhartha↗
NASA NTRS1995

On the utility of threads for data parallel programming

Threads provide a useful programming model for asynchronous behavior because of their ability to encapsulate units of work that can then be scheduled for execution at runtime, based on the dynamic state of a system. Recently, the threaded model has been applied to the domain of data parallel scientific codes, and initial reports indicate that the threaded model can produce performance gains over non-threaded approaches, primarily through the use of overlapping useful computation with communication latency. However, overlapping computation with communication is possible without the benefit of threads if the communication system supports asynchronous primitives, and this comparison has not been made in previous papers. This paper provides a critical look at the utility of lightweight threads as applied to data parallel scientific programming.

Fahringer, Thomas↗
DOE OSTI2026

Memory-Aware External Facelist Calculation: A Data-Parallel Atomic Hash Counting Approach

Unstructured volumetric meshes serve as fundamental data representations in various scientific simulations and analyses. They play a crucial role in representing complex computational domains and are essential for important numerical techniques, such as finite element analysis. Whenever such a mesh is read from a file, streamed in-situ, or generated by algorithms, scientific visualization libraries rely on calculating the external surface of a geometry, named “external facelist”, to produce a polygonal mesh for rendering. Consequently, external facelist calculation has become one of the most widely used algorithms in the scientific visualization domain, necessitating optimal performance. In this paper, we explore relevant work on external facelist calculation algorithms in two common visualization libraries, VTK and Viskores, assess their performance and memory constraints, and introduce a novel memory-aware external facelist calculation algorithm employing an atomic hash counting approach. This algorithm fully leverages Viskores' data-parallel primitive operations, facilitating its execution across diverse many-core architectures. Our algorithm features the lowest memory footprint on the GPU and the second-lowest on the CPU among all evaluated methods, and it also delivers the fastest performance on both CPU and GPU. It has been made available under an open-source license in the VTK and Viskores visualization systems.

Tsalikis, Spiros [Kitware] (ORCID:0000000151137195↗
NASA NTRS2009

Data Parallel Line Relaxation (DPLR) Code User Manual: Acadia - Version 4.01.1

Data-Parallel Line Relaxation (DPLR) code is a computational fluid dynamic (CFD) solver that was developed at NASA Ames Research Center to help mission support teams generate high-value predictive solutions for hypersonic flow field problems. The DPLR Code Package is an MPI-based, parallel, full three-dimensional Navier-Stokes CFD solver with generalized models for finite-rate reaction kinetics, thermal and chemical non-equilibrium, accurate high-temperature transport coefficients, and ionized flow physics incorporated into the code. DPLR also includes a large selection of generalized realistic surface boundary conditions and links to enable loose coupling with external thermal protection system (TPS) material response and shock layer radiation codes.

Wright, Michael J.↗
NASA NTRS1994

Runtime support for data parallel tasks

We have recently introduced a set of Fortran language extensions that allow for integrated support of task and data parallelism, and provide for shared data abstractions (SDA's) as a method for communications and synchronization among these tasks. In this paper we discuss the design and implementation issues of the runtime system necessary to support these extensions, and discuss the underlying requirements for such a system. To test the feasibility of this approach, we implement a prototype of the runtime system and use this to support an abstract multidisciplinary optimization (MDO) problem for aircraft design. We give initial results and discuss future plans.

Haines, Matthew↗
NASA NTRS1979

Codes for a priority queue on a parallel data bus

Some codes for arbitration of priorities among subsystem computers or peripheral device controllers connected to a parallel data bus are described. At arbitration time, several subsystems present wire-OR, parallel code words to the bus, and the central computer can identify the subsystem of highest priority and determine which of two or more transmission services the subsystem requires. A mathematical discussion of the optimality of the codes with regard to the number of subsystems that may participate in the scheme for a given number of wires is presented along with the number of services that each subsystem may request.

Wallis, D. E.↗
DOE OSTI2018

Data-parallel Python for High Energy Physics Analyses

In this paper, we explore features available in Python which are useful for data reduction tasks in High Energy Physics (HEP). Highlevel abstractions in Python are convenient for implementing data reduction tasks. However, in order for such abstractions to be practical, the efficiency of their performance must also be high. Because the data sets we process are typically large, we care about both I/O performance and in-memory processing speed. In particular, we evaluate the use of data-parallel programming, using MPI and numpy, to process a large experimental data set (42 TiB) stored in an HDF5 file. We measure the speed of processing of the data, distinguishing between the time spent reading data and the time spent processing the data in memory, and demonstrate the scalability of both, using up to 1200 KNL nodes (76800 cores) on Cori at NERSC

Paterno, Marc↗
DOE OSTI2020

Predicting Execution Times for Disk-based and In-Situ Parallel Data Analytics (Final Technical Report)

In recent years, there has been a significant amount of interests in in-situ analytics on simulation programs. For a variety of reasons, it is desirable to be able to predict the execution time of an analytics program. At the same time, frameworks such as MapReduce have become popular for scientific data analytics. This paper focuses on developing performance models for predicting execution time of parallel data analytics, with a special emphasis on in-situ analytics. We take two distinct approach towards performance prediction. We first expand SKOPE (a SKeleton framewOrk for Performance Exploration) with performance models for disk data read, cache performance, and page fault penalty. Second, an analytical performance model is also developed. We have evaluated our performance prediction framework as well as the analytical model on three hardware setups with well-known data mining algorithms implemented in three programming paradigms, MapReduce, MATE (a MapReduce-like parallel system with an alternate API for multi-core environments) and Smart (a MapReduce-like framework for in-situ analytics). Results show that our performance prediction framework along with the incorporated performance models are capable of accurately predicting execution times for parallel scientific analytics on different hardware setups.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2018

DPP-PMRF: Rethinking Optimization for a Probabilistic Graphical Model Using Data-Parallel Primitives

We present a new parallel algorithm for probabilistic graphical model optimization. The algorithm relies on data-parallel primitives (DPPs), which provide portable performance over hardware architecture. We evaluate results on CPUs and GPUs for an image segmentation problem. Compared to a serial baseline, we observe runtime speedups of up to 13X (CPU) and 44X (GPU). We also compare our performance to a reference, OpenMP-based algorithm, and find speedups of up to 7X (CPU).

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1993

Generating local addresses and communication sets for data-parallel programs

Generating local addresses and communication sets is an important issue in distributed-memory implementations of data-parallel languages such as High Performance FORTRAN. We show that, for an array A affinely aligned to a template that is distributed across p processors with a cyclic(k) distribution and a computation involving the regular section A(l:h:s), the local memory access sequence for any processor is characterized by a finite state machine of at most k states. We present fast algorithms for computing the essential information about these state machines, and extend the framework to handle multidimensional arrays. We also show how to generate communication sets using the state machine approach. Performance results show that this solution requires very little run-time overhead and acceptable preprocessing time.

Chatterjee, Siddhartha↗
NASA NTRS1993

Generating local addresses and communication sets for data-parallel programs

Generating local addresses and communication sets is an important issue in distributed-memory implementations of data-parallel languages such as High Performance Fortran. We show that for an array A affinely aligned to a template that is distributed across p processors with a cyclic(k) distribution, and a computation involving the regular section A, the local memory access sequence for any processor is characterized by a finite state machine of at most k states. We present fast algorithms for computing the essential information about these state machines, and extend the framework to handle multidimensional arrays. We also show how to generate communication sets using the state machine approach. Performance results show that this solution requires very little runtime overhead and acceptable preprocessing time.

Chatterjee, Siddhartha↗