Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Multicore”

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 181 records · Page 10

Parallelizable adjoint stencil computations using transposed forward-mode algorithmic differentiation

Algorithmic differentiation (AD) is a tool for generating discrete adjoint solvers, which efficiently compute gradients of functions with many inputs, for example for use in gradient-based optimization. AD is often applied to large computations such as stencil operators, which are an important part of most structured-mesh PDE solvers. Stencil computations are often parallelized, for example by using OpenMP, and optimized by using techniques such as cache-blocking and tiling to fully utilize multicore CPUs and many-core accelerators and GPUs. Differentiating these codes with conventional reverse-mode AD results in adjoint codes that cannot be expressed as stencil operations and may not be easily parallelizable. They thus leave most of the compute power of modern architectures unused. We present a method that combines forward-mode AD and loop transformation to generate adjoint solvers that use the same memory access pattern as the original computation that they are derived from and can benefit from the same optimization techniques. The effectiveness of this method is demonstrated by generating a scalable adjoint CFD solver for multicore CPUs and Xeon Phi accelerators.

algorithmic differentiation↗

Multi-core aware applications in CMS

One of the significant trends of recent years has been the move towards multicore CPU's with ever increasing numbers of cores. CMS has been preparing multicore aware applications that rely on 'multi-processing', namely the sharing of memory between processes forked from a single parent process. First experience with deploying these applications for production will be presented, as well as results from detailed profiling done to understand the limits of scaling with increasing numbers of cores.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

A Flexible-blocking Based Approach for Performance Tuning of Matrix Multiplication Routines for Large Matrices with Edge Cases

Efficient and scalable matrix operations are being highly demanding in the recent era of Machine Learning, Deep Learning, and Big Data Analytics. The two commonly used matrix-matrix operations in the Basic Linear Algebra Subprograms (BLAS) specification are General Matrix-Matrix multiplication (GEMM) and Symmetric Rank-k update (SYRK). The SYRK routine is a specialization of the GEMM routine, where half of the multiplications are skipped as the resultant matrix is known to be symmetric. Fortunately, several linear algebra libraries implement these BLAS routines quite efficiently. The libraries usually partition the input matrices into blocks and place them in processor caches, thus improving performance by leveraging the caches. However, the contemporary libraries are highly optimized for squarish matrices, but the performance degrades significantly for the matrices with edge case (strictly thin or strictly fat shapes) in the multicore machine. The primary reason is that the current state-of-the-art libraries make fixed block shapes based on a processor architecture, and do not consider the shape of the input matrices. In this paper, we propose a new blocking approach, we name it Flexible-blocking, to mitigate the scalability issues. In contrast to the contemporary libraries, our approach formulates the blocks of the input matrices based on the shapes of the matrices as well as the number of threads used in the implementation. Our proposed technique shows noticeable performance improvement on multicore shared-memory machines for the edge case matrices.

Hossain, Md Mosharaf↗

Roofline scaling trajectories: A method for parallel application and architectural performance analysis

The end of Dennard scaling signaled a shift in HPC supercomputer architectures from systems built from single-core processor architectures to systems built from multicore and eventually manycore architectures. This transition substantially complicated performance optimization and analysis as new programming models were created, new scaling methodologies deployed, and on-chip contention became a bottleneck to performance. Existing distributed memory performance models like logP and logGP were unable to capture this contention. The Roofline model was created to address this contention and its interplay with locality. However, to date, the Roofline model has focused on full-node concurrency. In this paper, we extend the Roofline model to capture the effects of concurrency on data locality and on-chip contention. We demonstrate the value of this new technique by evaluating the NAS parallel benchmarks on both multicore and manycore architectures under both strong-And weak-scaling regimes. In order to quantify the interplay between programming model and locality, we evaluate scaling under both the OpenMP and flat MPI programming models.

Ibrahim, K↗

Exploiting Data Similarity to Reduce Memory Footprints

Memory size has long limited large-scale applications on high-performance computing (HPC) systems. Since compute nodes frequently do not have swap space, physical memory often limits problem sizes. Increasing core counts per chip and power density constraints, which limit the number of DIMMs per node, have exacerbated this problem. Further, DRAM constitutes a significant portion of overall HPC system cost. Therefore, instead of adding more DRAM to the nodes, mechanisms to manage memory usage more efficiently - preferably transparently - could increase effective DRAM capacity and thus the benefit of multicore nodes for HPC systems. MPI application processes often exhibit significant data similarity. These data regions occupy multiple physical locations across the individual rank processes within a multicore node and thus offer a potential savings in memory capacity. These regions, primarily residing in heap, are dynamic, which makes them difficult to manage statically. Our novel memory allocation library, SBLLmalloc, automatically identifies identical memory blocks and merges them into a single copy. SBLLmalloc does not require application or OS changes since we implement it as a user-level library. Overall, we demonstrate that SBLLmalloc reduces the memory footprint of a range of MPI applications by 32.03% on average and up to 60.87%. Further, SBLLmalloc supports problem sizes for IRS over 21.36% larger than using standard memory management techniques, thus significantly increasing effective system size. Similarly, SBLLmalloc requires 43.75% fewer nodes than standard memory management techniques to solve an AMG problem.

97 MATHEMATICS AND COMPUTING↗

Real-Time Stochastic Optimization of Complex Energy Systems on High-Performance Computers

A scalable approach computes in operationally-compatible time the energy dispatch under uncertainty for electrical power grid systems of realistic size with thousands of scenarios. The authors propose several algorithmic and implementation advances in their parallel solver PIPS-IPM for stochastic optimization problems. New developments include a novel, incomplete, augmented, multicore, sparse factorization implemented within the PARDISO linear solver and new multicore- and GPU-based dense matrix implementations. They show improvement on the interprocess communication on Cray XK7 and XC30 systems. PIPS-IPM is used to solve 24-hour horizon power grid problems with up to 1.95 billion decision variables and 1.94 billion constraints on Cray XK7 and Cray XC30, with observed parallel efficiencies and solution times within an operationally defined time interval. To the authors' knowledge, "real-time"-compatible performance on a broad range of architectures for this class of problems hasn't been possible prior to this work

Computer Science↗

Strategies for Energy-Efficient Resource Management of Hybrid Programming Models

Many scientific applications are programmed using hybrid programming models that use both message passing and shared memory, due to the increasing prevalence of large-scale systems with multicore, multisocket nodes. Previous work has shown that energy efficiency can be improved using software-controlled execution schemes that consider both the programming model and the power-aware execution capabilities of the system. However, such approaches have focused on identifying optimal resource utilization for one programming model, either shared memory or message passing, in isolation. The potential solution space, thus the challenge, increases substantially when optimizing hybrid models since the possible resource configurations increase exponentially. Nonetheless, with the accelerating adoption of hybrid programming models, we increasingly need improved energy efficiency in hybrid parallel applications on large-scale systems. In this work, we present new software-controlled execution schemes that consider the effects of dynamic concurrency throttling (DCT) and dynamic voltage and frequency scaling (DVFS) in the context of hybrid programming models. Specifically, we present predictive models and novel algorithms based on statistical analysis that anticipate application power and time requirements under different concurrency and frequency configurations. Furthermore, we apply our models and methods to the NPB MZ benchmarks and selected applications from the ASC Sequoia codes. Overall, we achieve substantial energy savings (8.74 percent on average and up to 13.8 percent) with some performance gain (up to 7.5 percent) or negligible performance loss.

97 MATHEMATICS AND COMPUTING↗

A Programming Model Performance Study Using the NAS Parallel Benchmarks

Harnessing the power of multicore platforms is challenging due to the additional levels of parallelism present. In this paper we use the NAS Parallel Benchmarks to study three programming models, MPI, OpenMP and PGAS to understand their performance and memory usage characteristics on current multicore architectures. To understand these characteristics we use the Integrated Performance Monitoring tool and other ways to measure communication versus computation time, as well as the fraction of the run time spent in OpenMP. The benchmarks are run on two different Cray XT5 systems and an Infiniband cluster. Our results show that in general the three programming models exhibit very similar performance characteristics. In a few cases, OpenMP is significantly faster because it explicitly avoids communication. For these particular cases, we were able to re-write the UPC versions and achieve equal performance to OpenMP. Using OpenMP was also the most advantageous in terms of memory usage. Also we compare performance differences between the two Cray systems, which have quad-core and hex-core processors. We show that at scale the performance is almost always slower on the hex-core system because of increased contention for network resources.

Shan, Hongzhang↗

HPC Programming on Intel Many-Integrated-Core Hardware with MAGMA Port to Xeon Phi

This paper presents the design and implementation of several fundamental dense linear algebra (DLA) algorithms for multicore with Intel Xeon Phi coprocessors. In particular, we consider algorithms for solving linear systems. Further, we give an overview of the MAGMA MIC library, an open source, high performance library, that incorporates the developments presented here and, more broadly, provides the DLA functionality equivalent to that of the popular LAPACK library while targeting heterogeneous architectures that feature a mix of multicore CPUs and coprocessors. The LAPACK-compliance simplifies the use of the MAGMA MIC library in applications, while providing them with portably performant DLA. High performance is obtained through the use of the high-performance BLAS, hardware-specific tuning, and a hybridization methodology whereby we split the algorithm into computational tasks of various granularities. Execution of those tasks is properly scheduled over the heterogeneous hardware by minimizing data movements and mapping algorithmic requirements to the architectural strengths of the various heterogeneous hardware components. Our methodology and programming techniques are incorporated into the MAGMA MIC API, which abstracts the application developer from the specifics of the Xeon Phi architecture and is therefore applicable to algorithms beyond the scope of DLA.

97 MATHEMATICS AND COMPUTING↗

Parallel k-means++

A parallelization of the k-means++ seed selection algorithm on three distinct hardware platforms: GPU, multicore CPU, and multithreaded architecture. K-means++ was developed by David Arthur and Sergei Vassilvitskii in 2007 as an extension of the k-means data clustering technique. These algorithms allow people to cluster multidimensional data, by attempting to minimize the mean distance of data points within a cluster. K-means++ improved upon traditional k-means by using a more intelligent approach to selecting the initial seeds for the clustering process. While k-means++ has become a popular alternative to traditional k-means clustering, little work has been done to parallelize this technique. We have developed original C++ code for parallelizing the algorithm on three unique hardware architectures: GPU using NVidia's CUDA/Thrust framework, multicore CPU using OpenMP, and the Cray XMT multithreaded architecture. By parallelizing the process for these platforms, we are able to perform k-means++ clustering much more quickly than it could be done before.

Source record↗

Batched matrix computations on hardware accelerators based on GPUs

Scientific applications require solvers that work on many small size problems that are independent from each other. At the same time, the high-end hardware evolves rapidly and becomes ever more throughput-oriented and thus there is an increasing need for an effective approach to develop energy-efficient, high-performance codes for these small matrix problems that we call batched factorizations. The many applications that need this functionality could especially benefit from the use of GPUs, which currently are four to five times more energy efficient than multicore CPUs on important scientific workloads. This study, consequently, describes the development of the most common, one-sided factorizations, Cholesky, LU, and QR, for a set of small dense matrices. The algorithms we present together with their implementations are, by design, inherently parallel. In particular, our approach is based on representing the process as a sequence of batched BLAS routines that are executed entirely on a GPU. Importantly, this is unlike the LAPACK and the hybrid MAGMA factorization algorithms that work under drastically different assumptions of hardware design and efficiency of execution of the various computational kernels involved in the implementation. Thus, our approach is more efficient than what works for a combination of multicore CPUs and GPUs for the problems sizes of interest of the application use cases. The paradigm where upon a single chip (a GPU or a CPU) factorizes a single problem at a time is not at all efficient in our applications’ context. We illustrate all of these claims through a detailed performance analysis. With the help of profiling and tracing tools, we guide our development of batched factorizations to achieve up to two-fold speedup and three-fold better energy efficiency as compared against our highly optimized batched CPU implementations based on MKL library. Finally, the tested system featured two sockets of Intel Sandy Bridge CPUs and we compared with a batched LU factorizations featured in the CUBLAS library for GPUs, we achieve as high as 2.5× speedup on the NVIDIA K40 GPU.

97 MATHEMATICS AND COMPUTING↗

ConnectIt: a framework for static and incremental parallel graph connectivity algorithms

Connected components is a fundamental kernel in graph applications. The fastest existing multicore algorithms for solving graph connectivity are based on some form of edge sampling and/or linking and compressing trees. However, many combinations of these design choices have been left unexplored. In this paper, we design the ConnectIt framework, which provides different sampling strategies as well as various tree linking and compression schemes. ConnectIt enables us to obtain several hundred new variants of connectivity algorithms, most of which extend to computing spanning forest. In addition to static graphs, we also extend ConnectIt to support mixes of insertions and connectivity queries in the concurrent setting. We present an experimental evaluation of ConnectIt on a 72-core machine, which we believe is the most comprehensive evaluation of parallel connectivity algorithms to date. Compared to a collection of state-of-the-art static multicore algorithms, we obtain an average speedup of 12.4x (2.36x average speedup over the fastest existing implementation for each graph). Using ConnectIt, we are able to compute connectivity on the largest publicly-available graph (with over 3.5 billion vertices and 128 billion edges) in under 10 seconds using a 72-core machine, providing a 3.1x speedup over the fastest existing connectivity result for this graph, in any computational setting. For our incremental algorithms, we show that our algorithms can ingest graph updates at up to several billion edges per second. To guide the user in selecting the best variants in ConnectIt for different situations, we provide a detailed analysis of the different strategies. Finally, we show how the techniques in ConnectIt can be used to speed up two important graph applications: approximate minimum spanning forest and SCAN clustering.

Computer Science↗

Creating a Parallel Version of VisIt for Microsoft Windows

VisIt is a popular, free interactive parallel visualization and analysis tool for scientific data. Users can quickly generate visualizations from their data, animate them through time, manipulate them, and save the resulting images or movies for presentations. VisIt was designed from the ground up to work on many scales of computers from modest desktops up to massively parallel clusters. VisIt is comprised of a set of cooperating programs. All programs can be run locally or in client/server mode in which some run locally and some run remotely on compute clusters. The VisIt program most able to harness today's computing power is the VisIt compute engine. The compute engine is responsible for reading simulation data from disk, processing it, and sending results or images back to the VisIt viewer program. In a parallel environment, the compute engine runs several processes, coordinating using the Message Passing Interface (MPI) library. Each MPI process reads some subset of the scientific data and filters the data in various ways to create useful visualizations. By using MPI, VisIt has been able to scale well into the thousands of processors on large computers such as dawn and graph at LLNL. The advent of multicore CPU's has made parallelism the 'new' way to achieve increasing performance. With today's computers having at least 2 cores and in many cases up to 8 and beyond, it is more important than ever to deploy parallel software that can use that computing power not only on clusters but also on the desktop. We have created a parallel version of VisIt for Windows that uses Microsoft's MPI implementation (MSMPI) to process data in parallel on the Windows desktop as well as on a Windows HPC cluster running Microsoft Windows Server 2008. Initial desktop parallel support for Windows was deployed in VisIt 2.4.0. Windows HPC cluster support has been completed and will appear in the VisIt 2.5.0 release. We plan to continue supporting parallel VisIt on Windows so our users will be able to take full advantage of their multicore resources.

97 MATHEMATICS AND COMPUTING↗

A programming model performance study using the NAS parallel benchmarks

Harnessing the power of multicore platforms is challenging due to the additional levels of parallelism present. In this paper we use the NAS Parallel Benchmarks to study three programming models, MPI, OpenMP and PGAS to understand their performance and memory usage characteristics on current multicore architectures. To understand these characteristics we use the Integrated Performance Monitoring tool and other ways to measure communication versus computation time, as well as the fraction of the run time spent in OpenMP. The benchmarks are run on two different Cray XT5 systems and an Infiniband cluster. Our results show that in general the three programming models exhibit very similar performance characteristics. In a few cases, OpenMP is significantly faster because it explicitly avoids communication. For these particular cases, we were able to re-write the UPC versions and achieve equal performance to OpenMP. Using OpenMP was also the most advantageous in terms of memory usage. Also we compare performance differences between the two Cray systems, which have quad-core and hex-core processors.We show that at scale the performance is almost always slower on the hex-core system because of increased contention for network resources.

97 MATHEMATICS AND COMPUTING↗

SequenceL: Automated Parallel Algorithms Derived from CSP-NT Computational Laws

With the introduction of new parallel architectures like the cell and multicore chips from IBM, Intel, AMD, and ARM, as well as the petascale processing available for highend computing, a larger number of programmers will need to write parallel codes. Adding the parallel control structure to the sequence, selection, and iterative control constructs increases the complexity of code development, which often results in increased development costs and decreased reliability. SequenceL is a high-level programming language that is, a programming language that is closer to a human s way of thinking than to a machine s. Historically, high-level languages have resulted in decreased development costs and increased reliability, at the expense of performance. In recent applications at JSC and in industry, SequenceL has demonstrated the usual advantages of high-level programming in terms of low cost and high reliability. SequenceL programs, however, have run at speeds typically comparable with, and in many cases faster than, their counterparts written in C and C++ when run on single-core processors. Moreover, SequenceL is able to generate parallel executables automatically for multicore hardware, gaining parallel speedups without any extra effort from the programmer beyond what is required to write the sequen tial/singlecore code. A SequenceL-to-C++ translator has been developed that automatically renders readable multithreaded C++ from a combination of a SequenceL program and sample data input. The SequenceL language is based on two fundamental computational laws, Consume-Simplify- Produce (CSP) and Normalize-Trans - pose (NT), which enable it to automate the creation of parallel algorithms from high-level code that has no annotations of parallelism whatsoever. In our anecdotal experience, SequenceL development has been in every case less costly than development of the same algorithm in sequential (that is, single-core, single process) C or C++, and an order of magnitude less costly than development of comparable parallel code. Moreover, SequenceL not only automatically parallelizes the code, but since it is based on CSP-NT, it is provably race free, thus eliminating the largest quality challenge the parallelized software developer faces.

Cooke, Daniel↗

Mobile Thread Task Manager

The Mobile Thread Task Manager (MTTM) is being applied to parallelizing existing flight software to understand the benefits and to develop new techniques and architectural concepts for adapting software to multicore architectures. It allocates and load-balances tasks for a group of threads that migrate across processors to improve cache performance. In order to balance-load across threads, the MTTM augments a basic map-reduce strategy to draw jobs from a global queue. In a multicore processor, memory may be "homed" to the cache of a specific processor and must be accessed from that processor. The MTTB architecture wraps access to data with thread management to move threads to the home processor for that data so that the computation follows the data in an attempt to avoid L2 cache misses. Cache homing is also handled by a memory manager that translates identifiers to processor IDs where the data will be homed (according to rules defined by the user). The user can also specify the number of threads and processors separately, which is important for tuning performance for different patterns of computation and memory access. MTTM efficiently processes tasks in parallel on a multiprocessor computer. It also provides an interface to make it easier to adapt existing software to a multiprocessor environment.

Clement, Bradley J.↗

High-Performance Computing Optimization for Aladyn – Adaptive Neural Network Molecular Dynamics Mini-Application

This report provides a description and performance evaluation of the optimization techniques for high performance computing (HPC) implementation of the open source Computational Materials mini-application Aladyn (https://github.com/nasa/aladyn). Aladyn is a basic molecular dynamics code written in FORTRAN 2003, which is designed to demonstrate the use of adaptive neural networks (ANNs) in atomistic simulations. The role of ANNs is to efficiently reproduce the very complex energy landscape resulting from the atomic interactions in materials with the accuracy of the more expensive quantum mechanics-based calculations. The ANN is trained on a large set of atomic structures calculated using the density functional theory (DFT) method. While achieving orders of magnitude faster computational performance than DFT, the ANN-based approach was still very computationally demanding compared to the conventional approach of using empirically fitted energy functions. After its initial development, Aladyn was evaluated and optimized by experts at the NASA Advanced Supercomputing (NAS) division to exploit modern supercomputer architectures. The code has been optimized for execution on multicore central processing units (CPUs), including Intel® Skylake microarchitecture, and on graphic accelerators, such as Nvidia® V100 graphic processing units (GPUs), using Open Multi-Processing (OpenMP) and Open Accelerators (OpenACC) programming interfaces. The optimization achieved a speedup of 4.7 times the baseline version on CPU performance and an additional 2.4 times on CPU+GPU performance. Atomistic computer simulations are a fundamental tool in materials research to model material properties form physics-based first principles. Atomic interaction, governed by Quantum Mechanics (QM) require sophisticated and highly computationally demanding mathematical models to calculate [1]. Classical methods use approximate functional forms, empirically fitted through a set of variable parameters to emulate atomic energies as direct functions of atomic coordinates [2]. While empirical potentials are computationally much simpler, allowing simulations of large-scale systems of up to a trillion (1012) atoms [3], they are substantially less accurate compared to quantum calculations and applicable only to very specific atomic configurations or predefined crystallographic phases. A recently suggested approach is to use heuristic machine learning methods [4], such as those based on Adaptive Neural Networks (ANNs) to predict atomic energies, after being trained on a sufficiently large database of QM-calculated structures [5,6]. This approach reduces significantly the computational complexity, allowing for simulations of orders of magnitude larger systems compared to QM-based methods without compromising accuracy. Still, compared to classical methods using empirical energy functions, ANN methods remain two- to three orders of magnitude more computationally demanding. Hence, the computational cost of simulations, together with the need for extensive training of ANNs, still makes the practical implementation of ANN-based methods quite challenging. The purpose of the Aladyn mini-application software [7], available as open source at https://github.com/nasa/aladyn, is to be a testbed for exploring possible optimization strategies to develop highly scalable parallel algorithms for ANN-based atomistic simulations. Aladyn is aimed at utilizing the architecture of the high-end modern highperformance computing (HPC) hardware based on multicore central processing units (CPUs) equipped with graphic processing unit (GPU) accelerators. Specifically, the goal is to optimize the performance on a single HPC compute node, before implementing scaling to multi-node parallelization using message passing interface (MPI). At the same time, the open source code of Aladyn can serve as a training model for students and professors in academia.

Yamakov, Vesselin I.↗

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↗