Search NASA⌕ Search

SEARCH · Search NASA

Results for “memory access optimization”

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 127 records · Page 7

NVMalloc: Exposing an Aggregate SSD Store as a Memory Partition in Extreme-Scale Machines

DRAM is a precious resource in extreme-scale machines and is increasingly becoming scarce, mainly due to the growing number of cores per node. On future multi-petaflop and exaflop machines, the memory pressure is likely to be so severe that we need to rethink our memory usage models. Fortunately, the advent of non-volatile memory (NVM) offers a unique opportunity in this space. Current NVM offerings possess several desirable properties, such as low cost and power efficiency, but suffer from high latency and lifetime issues. We need rich techniques to be able to use them alongside DRAM. In this talk, we present a novel approach for exploiting NVM as a secondary memory partition so that applications can explicitly allocate and manipulate memory regions therein. More specifically, we propose an NVMalloc library with a suite of services that enables applications to access a distributed NVM storage system. We have devised ways within NVMalloc so that the storage system, built from compute node-local NVM devices, can be accessed in a byte-addressable fashion using the memory mapped I/O interface. Our approach has the potential to re-energize out-of-core computations on largescale machines by having applications allocate certain variables through NVMalloc, thereby increasing the overall memory capacity available. Our evaluation on a 128-core cluster shows that NVMalloc enables applications to compute problem sizes larger than the physical memory in a cost-effective manner. It can bring more performance/efficiency gain with increased computation time between NVM memory accesses or increased data access locality. In addition, our results suggest that while NVMalloc enables transparent access to NVM-resident variables, the explicit control it provides is crucial to optimize application performance.

99 GENERAL AND MISCELLANEOUS↗

An implementation and evaluation of the MPI 3.0 one-sided communication interface

The Q1 Message Passing Interface (MPI) 3.0 standard includes a significant revision to MPI’s remote memory access (RMA) interface, which provides support for one-sided communication. MPI-3 RMA is expected to greatly enhance the usability and performance ofMPI RMA.We present the first complete implementation of MPI-3 RMA and document implementation techniques and performance optimization opportunities enabled by the new interface. Our implementation targets messaging-based networks and is publicly available in the latest release of the MPICH MPI implementation. Here using this implementation, we explore the performance impact of new MPI-3 functionality and semantics. Results indicate that the MPI-3 RMA interface provides significant advantages over the MPI-2 interface by enabling increased communication concurrency through relaxed semantics in the interface and additional routines that provide new window types, synchronization modes, and atomic operations.

97 MATHEMATICS AND COMPUTING↗

Performance Modeling and Measurement of Parallelized Code for Distributed Shared Memory Multiprocessors

This paper presents a model to evaluate the performance and overhead of parallelizing sequential code using compiler directives for multiprocessing on distributed shared memory (DSM) systems. With increasing popularity of shared address space architectures, it is essential to understand their performance impact on programs that benefit from shared memory multiprocessing. We present a simple model to characterize the performance of programs that are parallelized using compiler directives for shared memory multiprocessing. We parallelized the sequential implementation of NAS benchmarks using native Fortran77 compiler directives for an Origin2000, which is a DSM system based on a cache-coherent Non Uniform Memory Access (ccNUMA) architecture. We report measurement based performance of these parallelized benchmarks from four perspectives: efficacy of parallelization process; scalability; parallelization overhead; and comparison with hand-parallelized and -optimized version of the same benchmarks. Our results indicate that sequential programs can conveniently be parallelized for DSM systems using compiler directives but realizing performance gains as predicted by the performance model depends primarily on minimizing architecture-specific data locality overhead.

Waheed, Abdul↗

Gregarious Data Re-structuring in a Many Core Architecture

this paper, we have developed a new methodology that takes in consideration the access patterns from a single parallel actor (e.g. a thread), as well as, the access patterns of “grouped” parallel actors that share a resource (e.g. a distributed Level 3 cache). We start with a hierarchical tile code for our target machine and apply a series of transformations at the tile level to improve data residence in a given memory hierarchy level. The contribution of this paper includes (a) collaborative data restructuring for group reuse and (b) low overhead transformation technique to improve access pattern and bring closely connected data elements together. Preliminary results in a many core architecture, Tilera TileGX, shows promising improvements over optimized OpenMP code (up to 31% increase in GFLOPS) and over our own previous work on fine grained runtimes (up to 16%) for selected kernels

data restructuring↗

Retention modeling for ultra-thin density of Cu-based conductive bridge random access memory (CBRAM)

We investigate the effect of Cu concentration On-state resistance retention characteristics of W/Cu/Ti/HfO{sub 2}/Pt memory cell. The development of RRAM device for application depends on the understanding of the failure mechanism and the key parameters for device optimization. In this study, we develop analytical expression for cations (Cu{sup +}) diffusion model using Gaussian distribution for detailed analysis of data retention time at high temperature. It is found that the improvement of data retention time depends not only on the conductive filament (CF) size but also on Cu atoms concentration density in the CF. Based on the simulation result, better data retention time is observed for electron wave function associated with Cu{sup +} overlap and an extended state formation. This can be verified by analytical calculation of Cu atom defects inside the filament, based on Cu{sup +} diffusion model. The importance of Cu diffusion for the device reliability and the corresponding local temperature of the filament were analyzed by COMSOL Multiphysics simulation.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

CoREC: Scalable and Resilient In-memory Data Staging for In-situ Workflows

The dramatic increase in the scale of current and planned high-end HPC systems is leading new challenges, such as the growing costs of data movement and IO, and the reduced mean time between failures (MTBF) of system components. In-situ workflows, i.e., executing the entire application workflows on the HPC system, have emerged as an attractive approach to address data-related challenges by moving computations closer to the data, and staging-based frameworks have been effectively used to support in-situ workflows at scale. However, the resilience of these staging-based solutions has not been addressed, and they remain susceptible to expensive data failures. Furthermore, naive use of data resilience techniques such as n-way replication and erasure codes can impact latency and/or result in significant storage overheads. In this article, we present CoREC, a scalable and resilient in-memory data staging runtime for large-scale in-situ workflows. CoREC uses a novel hybrid approach that combines dynamic replication with erasure coding based on data access patterns. It also leverages multiple levels of replications and erasure coding to support diverse data resiliency requirements. Furthermore, the article presents optimizations for load balancing and conflict-avoiding encoding, and a low overhead, lazy data recovery scheme. We have implemented the CoREC runtime and have deployed with the DataSpaces staging service on leadership class computing machines and present an experimental evaluation in the article. Here, the experiments demonstrate that CoREC can tolerate in-memory data failures while maintaining low latency and sustaining high overall storage efficiency at large scales.

97 MATHEMATICS AND COMPUTING↗

Towards Scalable Deep Learning via I/O Analysis and Optimization

Deep learning systems have been growing in prominence as a way to automatically characterize objects, trends, and anomalies. Given the importance of deep learning systems, researchers have been investigating techniques to optimize such systems. An area of particular interest has been using large supercomputing systems to quickly generate effective deep learning networks: a phase often referred to as “training” of the deep learning neural network. As we scale existing deep learning frameworks—such as Caffe—on these large supercomputing systems, we notice that the parallelism can help improve the computation tremendously, leaving data I/O as the major bottleneck limiting the overall system scalability. In this paper, we first present a detailed analysis of the performance bottlenecks of Caffe on large supercomputing systems. Our analysis shows that the I/O subsystem of Caffe—LMDB—relies on memory-mapped I/O to access its database, which can be highly inefficient on large-scale systems because of its interaction with the process scheduling system and the network-based parallel filesystem. Based on this analysis, we then present LMDBIO, our optimized I/O plugin for Caffe that takes into account the data access pattern of Caffe in order to vastly improve I/O performance. Our experimental results show that LMDBIO can improve the overall execution time of Caffe by nearly 20-fold in some cases.

Pumma, Sarunya↗

Global Trees: A Framework for Linked Data Structures on Distributed Memory Parallel Systems

This paper describes the Global Trees (GT) system that provides a multi-layered interface to a global address space view of distributed tree data structures, while providing scalable performance on distributed memory systems. The Global Trees system utilizes coarse-grained data movement to enhance locality and communication efficiency. We describe the design and implementation of GT, illustrate its use in the context of a gravitational simulation application, and provide experimental results that demonstrate the effectiveness of the approach. The key benefits of using this system include efficient sharedmemory style programming of distributed trees, tree-specific optimizations for data access and computation, and the ability to customize many aspects of GT to optimize application performance.

97 MATHEMATICS AND COMPUTING↗

TU-AB-BRC-10: Modeling of Radiotherapy Linac Source Terms Using ARCHER Monte Carlo Code: Performance Comparison of GPU and MIC Computing Accelerators

Purpose: (1) To perform phase space (PS) based source modeling for Tomotherapy and Varian TrueBeam 6 MV Linacs, (2) to examine the accuracy and performance of the ARCHER Monte Carlo code on a heterogeneous computing platform with Many Integrated Core coprocessors (MIC, aka Xeon Phi) and GPUs, and (3) to explore the software micro-optimization methods. Methods: The patient-specific source of Tomotherapy and Varian TrueBeam Linacs was modeled using the PS approach. For the helical Tomotherapy case, the PS data were calculated in our previous study (Su et al. 2014 41(7) Medical Physics). For the single-view Varian TrueBeam case, we analytically derived them from the raw patient-independent PS data in IAEA’s database, partial geometry information of the jaw and MLC as well as the fluence map. The phantom was generated from DICOM images. The Monte Carlo simulation was performed by ARCHER-MIC and GPU codes, which were benchmarked against a modified parallel DPM code. Software micro-optimization was systematically conducted, and was focused on SIMD vectorization of tight for-loops and data prefetch, with the ultimate goal of increasing 512-bit register utilization and reducing memory access latency. Results: Dose calculation was performed for two clinical cases, a Tomotherapy-based prostate cancer treatment and a TrueBeam-based left breast treatment. ARCHER was verified against the DPM code. The statistical uncertainty of the dose to the PTV was less than 1%. Using double-precision, the total wall time of the multithreaded CPU code on a X5650 CPU was 339 seconds for the Tomotherapy case and 131 seconds for the TrueBeam, while on 3 5110P MICs it was reduced to 79 and 59 seconds, respectively. The single-precision GPU code on a K40 GPU took 45 seconds for the Tomotherapy dose calculation. Conclusion: We have extended ARCHER, the MIC and GPU-based Monte Carlo dose engine to Tomotherapy and Truebeam dose calculations.

60 APPLIED LIFE SCIENCES↗

Extreme-Scale Algorithms & Software Resilience (EASIR) Architecture-Aware Algorithms for Scalable Performance and Resilience on Heterogeneous Architectures

This project addresses both communication-avoiding algorithms, and reproducible floating-point computation. Communication, i.e. moving data, either between levels of memory or processors over a network, is much more expensive per operation than arithmetic (measured in time or energy), so we seek algorithms that greatly reduce communication. We developed many new algorithms for both dense and sparse, and both direct and iterative linear algebra, attaining new communication lower bounds, and getting large speedups in many cases. We also extended this work in several ways: (1) We minimize writes separately from reads, since writes may be much more expensive than reads on emerging memory technologies, like Flash, sometimes doing asymptotically fewer writes than reads. (2) We extend the lower bounds and optimal algorithms to arbitrary algorithms that may be expressed as perfectly nested loops accessing arrays, where the array subscripts may be arbitrary affine functions of the loop indices (eg A(i), B(i,j+k, k+3*m-7, …) etc.). (3) We extend our communication-avoiding approach to some machine learning algorithms, such as support vector machines. This work has won a number of awards. We also address reproducible floating-point computation. We define reproducibility to mean getting bitwise identical results from multiple runs of the same program, perhaps with different hardware resources or other changes that should ideally not change the answer. Many users depend on reproducibility for debugging or correctness. However, dynamic scheduling of parallel computing resources, combined with nonassociativity of floating point addition, makes attaining reproducibility a challenge even for simple operations like summing a vector of numbers, or more complicated operations like the Basic Linear Algebra Subprograms (BLAS). We describe an algorithm that computes a reproducible sum of floating point numbers, independent of the order of summation. The algorithm depends only on a subset of the IEEE Floating Point Standard 754-2008, uses just 6 words to represent a “reproducible accumulator,” and requires just one read-only pass over the data, or one reduction in parallel. New instructions based on this work are being considered for inclusion in the future IEEE 754-2018 floating-point standard, and new reproducible BLAS are being considered for the next version of the BLAS standard.

97 MATHEMATICS AND COMPUTING↗

Digital correlation detector for low-cost Omega navigation

Techniques to lower the cost of using the Omega global navigation network with phase-locked loops (PLL) were developed. The technique that was accepted as being "optimal" is called the memory-aided phase-locked loop (MAPLL) since it allows operation on all eight Omega time slots with one PLL through the implementation of a random access memory. The receiver front-end and the signals that it transmits to the PLL were first described. A brief statistical analysis of these signals was then made to allow a rough comparison between the front-end presented in this work and a commercially available front-end to be made. The hardware and theory of application of the MAPLL were described, ending with an analysis of data taken with the MAPLL. Some conclusions and recommendations were also given.

Chamberlin, K. A.↗

Data traffic reduction schemes for sparse Cholesky factorizations

Load distribution schemes are presented which minimize the total data traffic in the Cholesky factorization of dense and sparse, symmetric, positive definite matrices on multiprocessor systems with local and shared memory. The total data traffic in factoring an n x n sparse, symmetric, positive definite matrix representing an n-vertex regular 2-D grid graph using n (sup alpha), alpha is equal to or less than 1, processors are shown to be O(n(sup 1 + alpha/2)). It is O(n(sup 3/2)), when n (sup alpha), alpha is equal to or greater than 1, processors are used. Under the conditions of uniform load distribution, these results are shown to be asymptotically optimal. The schemes allow efficient use of up to O(n) processors before the total data traffic reaches the maximum value of O(n(sup 3/2)). The partitioning employed within the scheme, allows a better utilization of the data accessed from shared memory than those of previously published methods.

Naik, Vijay K.↗

Implementing Arbitrary/Common Concurrent Writes of CRCW PRAM

The Parallel Random Access Machines (PRAM) abstraction is the simplest and most elegant algorithmic model for the design and analysis of parallel algorithms. It consists of different models categorized based on the underlying memory access mode used, the most powerful of which is the Concurrent Read Concurrent Write (CRCW) model. A PRAM algorithm describes a series of rounds, each of which consists of a collection of operations that can be executed concurrently within the same time step. However, the lack of support for concurrent memory accesses and the prevalence of asynchronous programming models led to the belief that implementing CRCW PRAM algorithms is unattainable and prompted many to avoid this model except for theoretical studies of optimal performance.In this work, we study the arbitrary and common concurrent writes in the CRCW PRAM model and explore implementation challenges on general-purpose systems. Moreover, we examine current practices for implementing common/arbitrary concurrent writes and propose a new efficient lightweight and thread-safe method to implement concurrent writes through leveraging atomic instructions. To demonstrate the efficacy of our method, we developed OpenMP kernels for classical CRCW PRAM algorithms and provide experimental results and comparisons based on run time performance measured over the x86 multicore architecture. Our results show a performance speedup compared to current practices up to 4.5x across all our benchmarks.

Ghanim, Fady↗

ZrO2 Layer Thickness Dependent Electrical and Dielectric Properties of BST/ZrO2/BST Multilayer Thin Films

Recently, high K materials play an important role in microelectronic devices such as capacitors, memory devices, and microwave devices. Now a days ferroelectric barium strontium titanate [Ba{sub x}Sr{sub 1-x}TiO{sub 3}, (BST)] thin film is being actively investigated for applications in dynamic random access memories (DRAM), field effect transistor (FET), and tunable devices because of its properties such as high dielectric constant, low leakage current, low dielectric loss, and high dielectric breakdown strength. Several approaches have been used to optimize the dielectric and electrical properties of BST thin films such as doping, graded compositions, and multilayer structures. We have found that inserting a ZrO{sub 2} layer in between two BST layers results in a significant reduction in dielectric constant, loss tangent, and leakage current in the multilayer thin films. Also it is shown that the properties of multilayer structure are found to depend strongly on the sublayer thicknesses. In this work the effect of ZrO{sub 2} layer thickness on the dielectric, ferroelectric as well as electrical properties of BST/ZrO{sub 2}/BST multilayer structure is studied. The multilayer Ba{sub 0.8}Sr{sub 0.2}TiO{sub 3}/ZrO{sub 2}/Ba{sub 0.8}Sr{sub 0.2}TiO{sub 3} film is deposited by a sol-gel process on the platinized Si substrate. The thickness of the middle ZrO{sub 2} layer is varied while keeping the top and bottom BST layer thickness as fixed. It is observed that the dielectric constant, dielectric loss tangent, and leakage current of the multilayer films reduce with the increase of ZrO{sub 2} layer thickness and hence suitable for memory device applications. The ferroelectric properties of the multilayer film also decrease with the ZrO{sub 2} layer thickness.

14 SOLAR ENERGY↗

Helium: lifting high-performance stencil kernels from stripped x86 binaries to halide DSL code

Highly optimized programs are prone to bit rot, where performance quickly becomes suboptimal in the face of new hardware and compiler techniques. In this paper we show how to automatically lift performance-critical stencil kernels from a stripped x86 binary and generate the corresponding code in the high-level domain-specific language Halide. Using Halide's state-of-the-art optimizations targeting current hardware, we show that new optimized versions of these kernels can replace the originals to rejuvenate the application for newer hardware. The original optimized code for kernels in stripped binaries is nearly impossible to analyze statically. Instead, we rely on dynamic traces to regenerate the kernels. We perform buffer structure reconstruction to identify input, intermediate and output buffer shapes. Here, we abstract from a forest of concrete dependency trees which contain absolute memory addresses to symbolic trees suitable for high-level code generation. This is done by canonicalizing trees, clustering them based on structure, inferring higher-dimensional buffer accesses and finally by solving a set of linear equations based on buffer accesses to lift them up to simple, high-level expressions. Helium can handle highly optimized, complex stencil kernels with input-dependent conditionals. We lift seven kernels from Adobe Photoshop giving a 75 % performance improvement, four kernels from Irfan View, leading to 4.97 x performance, and one stencil from the mini GMG multigrid benchmark netting a 4.25 x improvement in performance. We manually rejuvenated Photoshop by replacing eleven of Photoshop's filters with our lifted implementations, giving 1.12 x speedup without affecting the user experience.

97 MATHEMATICS AND COMPUTING↗

Code modernization strategies for short-range non-bonded molecular dynamics simulations

Modern HPC systems are increasingly relying on greater core counts and wider vector registers. Thus, applications need to be adapted to fully utilize these hardware capabilities. One class of applications that can benefit from this increase in parallelism are molecular dynamics simulations. In this paper, we describe our efforts at modernizing the ESPResSo++ simulation package for molecular dynamics by restructuring its particle data layout for efficient memory accesses and applying vectorization techniques to benefit the calculation of short-range non-bonded forces, which results in an overall three times speedup and serves as a baseline for further optimizations. We also implement fine-grained parallelism for multi-core CPUs through HPX, a C++ runtime system which uses lightweight threads and an asynchronous many-task approach to maximize concurrency. Our goal is to evaluate the performance of an HPX-based approach compared to the bulk-synchronous MPI-based implementation. This requires the introduction of an additional layer to the domain decomposition scheme that defines the task granularity. On spatially inhomogeneous systems, which impose a corresponding load-imbalance in traditional MPI-based approaches, we demonstrate that by choosing an optimal task size, the efficient work-stealing mechanisms of HPX can overcome the overhead of communication resulting in an overall 1.4 times speedup compared to the baseline MPI version.

97 MATHEMATICS AND COMPUTING↗

Instruction Roofline: An insightful visual performance model for GPUs

The Roofline performance model provides an intuitive approach to identify performance bottlenecks and guide performance optimization. However, the classic FLOP-centric approach is inappropriate for the emerging applications that perform more integer operations than floating point operations. In this article, we reintroduce our Instruction Roofline Model on NVIDIA GPUs and expand our evaluation of it. The Instruction Roofline incorporates instructions and memory transactions across all memory hierarchies together, and provides more performance insights than the FLOP-oriented Roofline Model, that is, instruction throughput, stride memory access patterns, bank conflicts, and thread predication. We use our Instruction Roofline methodology to analyze eight proxy applications: HPGMG from AMReX, Matrix Transpose benchmarks, ADEPT from MetaHipMer's sequence alignment phase, EXTENSION from MetaHipMer's local assembly phase, CUSP, cuSPARSE, cudaTensorCoreGemm, and cuBLAS. We demonstrate the ability of our methodology to understand various aspects of performance and performance bottlenecks on NVIDIA GPUs and motivate code optimizations.

Ding, N↗

Scaling Semantic Graph Databases in Size and Performance

In this paper we present SGEM, a full software system for accelerating large-scale semantic graph databases on commodity clusters. Unlike current approaches, SGEM addresses semantic graph databases by only employing graph methods at all the levels of the stack. On one hand, this allows exploiting the space efficiency of graph data structures and the inherent parallelism of graph algorithms. These features adapt well to the increasing system memory and core counts of modern commodity clusters. On the other hand, however, these systems are optimized for regular computation and batched data transfers, while graph methods usually are irregular and generate fine-grained data accesses with poor spatial and temporal locality. Our framework comprises a SPARQL to data parallel C compiler, a library of parallel graph methods and a custom, multithreaded runtime system. We introduce our stack, motivate its advantages with respect to other solutions and show how we solved the challenges posed by irregular behaviors. We present the result of our software stack on the Berlin SPARQL benchmarks with datasets up to 10 billion triples (a triple corresponds to a graph edge), demonstrating scaling in dataset size and in performance as more nodes are added to the cluster.

Aggregation↗