Search NASA⌕ Search

SEARCH · Search NASA

Results for “data movement”

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 217 records · Page 12

Distributed-Memory Sparse Deep Neural Network Inference Using Global Arrays

Partitioned Global Address Space (PGAS) models exhibit tremendous promise in developing efficient and productive distributed-memory parallel applications. They have been used extensively in scientific computations due to conveniently offering a ``shared-memory''-like model and convenient interfaces that separate communication with synchronization. Traditionally, PGAS communication models have been applied to dense/contiguously distributed data, but most modern applications depict varied levels of sparsity. Existing PGAS models require certain adaptations to support distributed sparse computations, since associated computations often require matrix arithmetic, in addition to data movement. The Global Arrays toolkit from Pacific Northwest National Laboratory (PNNL) is one of the earliest PGAS models to combine one-sided data communication and distributed matrix operations and is still used in the popular NWChem quantum chemistry suite. Recently, we have expanded the Global Arrays toolkit to support common sparse operations, like sparse matrix-dense matrix multiplies (SpMM), sparse matrix-sparse matrix multiplication (SpGEMM) and Sampled Dense-Dense Matrix Multiplication (SDDMM). As it turns out, these operations are the bedrock of sparse Deep Learning (DL); sparse deep neural networks and Graph Neural Networks (GNNs) have gained increasing attention recently in achieving speedups on training and inference with reduced memory footprints. Unlike scientific applications in High Performance Computing (HPC), modern (distributed-memory capable) DL toolkits often rely on non-standardized and closed-source vendor software optimizations, creating challenges in software-hardware co-design at scale. Our goal is to support a variety of distributed-memory sparse matrix operations and helper functions in the newly created Sparse Global Arrays (SGA), such that it is possible to build portable and productive Machine Learning scenarios for algorithm/software and hardware codesign purposes. Contemporary data-parallel schemes for training/inference are undergoing a major overhaul since model replication limits scalability and causes resource inefficiencies. As such, we have adopted tensor parallelism in decomposing the model and inputs, to mitigate memory issues. Current implementation is built on top of MPI and uses CPUs to maximize the portability across the platforms.

Distributed computing, machine learning↗

Slycat™ User Manual

Slycat™ is a web-based system for performing data analysis and visualization of potentially large quantities of remote, high-dimensional data. Slycat™ specializes in working with ensemble data. An ensemble is a group of related data sets, which typically consists of a set of simulation runs exploring the same problem space. An ensemble can be thought of as a set of samples within a multi-variate domain, where each sample is a vector whose value defines a point in high-dimensional space. To understand and describe the underlying problem being modeled in the simulations, ensemble analysis looks for shared behaviors and common features across the group of runs. Additionally, ensemble analysis tries to quantify differences found in any members that deviate from the rest of the group. The Slycat™ system integrates data management, scalable analysis, and visualization. Results are viewed remotely on a user’s desktop via commodity web clients using a multi-tiered hierarchy of computation and data storage, as shown in Figure 1. Our goal is to operate on data as close to the source as possible, thereby reducing time and storage costs associated with data movement. Consequently, we are working to develop parallel analysis capabilities that operate on High Performance Computing (HPC) platforms, to explore approaches for reducing data size, and to implement strategies for staging computation across the Slycat™ hierarchy. Within Slycat™, data and visual analysis are organized around projects, which are shared by a project team. Project members are explicitly added, each with a designated set of permissions. Although users sign-in to access Slycat™, individual accounts are not maintained. Instead, authentication is used to determine project access. Within projects, Slycat™ models capture analysis results and enable data exploration through various visual representations. Although for scientists each simulation run is a model of real-world phenomena given certain conditions, we use the term model to refer to our modeling of the ensemble data, not the physics. Different model types often provide complementary perspectives on data features when analyzing the same data set. Each model visualizes data at several levels of abstraction, allowing the user to range from viewing the ensemble holistically to accessing numeric parameter values for a single run. Bookmarks provide a mechanism for sharing results, enabling interesting model states to be labeled and saved.

97 MATHEMATICS AND COMPUTING↗

Arithmetic Data Cube as a Data Intensive Benchmark

Data movement across computational grids and across memory hierarchy of individual grid machines is known to be a limiting factor for application involving large data sets. In this paper we introduce the Data Cube Operator on an Arithmetic Data Set which we call Arithmetic Data Cube (ADC). We propose to use the ADC to benchmark grid capabilities to handle large distributed data sets. The ADC stresses all levels of grid memory by producing 2d views of an Arithmetic Data Set of d-tuples described by a small number of parameters. We control data intensity of the ADC by controlling the sizes of the views through choice of the tuple parameters.

Frumkin, Michael A.↗

Analyzing and Visualizing Cosmological Simulations with ParaView

The advent of large cosmological sky surveys - ushering in the era of precision cosmology - has been accompanied by ever larger cosmological simulations. The analysis of these simulations, which currently encompass tens of billions of particles and up to a trillion particles in the near future, is often as daunting as carrying out the simulations in the first place. Therefore, the development of very efficient analysis tools combining qualitative and quantitative capabilities is a matter of some urgency. In this paper, we introduce new analysis features implemented within ParaView, a fully parallel, open-source visualization toolkit, to analyze large N-body simulations. A major aspect of ParaView is that it can live and operate on the same machines and utilize the same parallel power as the simulation codes themselves. In addition, data movement is in a serious bottleneck now and will become even more of an issue in the future; an interactive visualization and analysis tool that can handle data in situ is fast becoming essential. The new features in ParaView include particle readers and a very efficient halo finder that identifies friends-of-friends halos and determines common halo properties, including spherical overdensity properties. In combination with many other functionalities already existing within ParaView, such as histogram routines or interfaces to programming languages like Python, this enhanced version enables fast, interactive, and convenient analyses of large cosmological simulations. In addition, development paths are available for future extensions.

79 ASTRONOMY AND ASTROPHYSICS↗

Combining In-situ and In-transit Processing to Enable Extreme-Scale Scientific Analysis

With the onset of extreme-scale computing, I/O constraints make it increasingly difficult for scientists to save a sufficient amount of raw simulation data to persistent storage. One potential solution is to change the data analysis pipeline from a post-process centric to a concurrent approach based on either in-situ or in-transit processing. In this context computations are considered in-situ if they utilize the primary compute resources, while in-transit processing refers to offloading computations toa set of secondary resources using asynchronous data transfers. In this paper we explore the design and implementation of three common analysis techniques typically performed on large-scale scientific simulations: topological analysis, descriptive statistics, and visualization. We summarize algorithmic developments, describe a resource scheduling system to coordinate the execution ofvarious analysis workflows, and discuss our implementation using the DataSpaces and ADIOS frameworks that support efficient data movement between in-situ and in-transit computations. Wedemonstrate the efficiency of our lightweight, flexible framework by deploying it on the Jaguar XK6 to analyze data generated by S3D, a massively parallel turbulent combustion code. Our framework allows scientists dealing with the data deluge at extreme scale to perform analyses at increased temporal resolutions, mitigate I/O costs, and significantly improve the time to insight.

Bennett, Janine C.↗

ASCR Workshop on In Situ Data Management: Enabling Scientific Discovery from Diverse Data Sources

In January 2019, the U.S. Department of Energy, Office of Science program in Advanced Scientific Computing Research, convened a workshop to identify priority research directions for in situ data management (ISDM). The workshop defined ISDM as the practices, capabilities, and procedures to control the organization of data and enable the coordination and communication among heterogeneous tasks, executing simultaneously in a high-performance computing system, cooperating toward a common objective. The workshop revealed two primary, interdependent motivations for processing and managing data in situ. The first motivation is that the in situ methodology enables scientific discovery from a broad range of data sources over a wide scale of computing platforms: leadership-class systems, clusters, clouds, workstations, and embedded devices at the edge. The successful development of ISDM capabilities will benefit real-time decision-making, design optimization, and data-driven scientific discovery. The second motivation is the need to decrease data volumes. ISDM can make critical contributions to managing large data volumes from computations and experiments to minimize data movement, save storage space, and boost resource efficiency, often while simultaneously increasing scientific precision. A fundamental finding of this workshop is that the methodologies used to manage data among a variety of tasks in situ can be used to facilitate scientific discovery from many different data sources—simulation, experiment, and sensors, for example—and that being able to do so at numerous computing scales will benefit real-time decision-making, design optimization, and data-driven scientific discovery across the Office of Science mission space. Applications wanting to use the in situ capabilities include those where data analysis feeds back to the simulation, decisions are made autonomously, big data or machine learning is among the tasks to be coordinated, and computations need to be completed in real time. The workshop identified six priority research directions that highlight the components and capabilities needed for ISDM to be successful for the wide variety of applications discussed: making ISDM capabilities more pervasive, controllable, composable, and transparent, with a focus on greater coordination with the software stack and a diversity of fundamentally new data algorithms.

97 MATHEMATICS AND COMPUTING↗

Patient Motion and Targeting Accuracy in Robotic Spinal Radiosurgery: 260 Single-Fraction Fiducial-Free Cases

Purpose: To evaluate clinical targeting precision and assess patient movement data during fiducial-free, single-fraction spinal radiosurgery with the Cyberknife (CK). Methods and Materials: Image-guided spine tracking accuracy was tested using two phantoms. Movement patterns (three translations, roll, pitch and yaw) were obtained from log files of 260 patient treatments (47 cervical, 89 thoracic, 90 lumbar, and 34 pelvic/sacral). For two treatments (average and maximum motion scenario), we added offsets to all beams according to recorded patient movements and recalculated the delivered dose distribution to simulate the dosimetric impact of intrafraction motion. Results: Phantom spine position was registered with an accuracy of <0.2 mm for translational and <0.3{sup o} for rotational directions. Residual patient motion yielded mean targeting errors per beam of 0.28 {+-} 0.13 mm (X), 0.25 {+-} 0.15 mm (Y), 0.19 {+-} 0.11 mm (Z) and 0.40 {+-} 0.20{sup o} (roll), 0.20 {+-} 0.08{sup o} (pitch), and 0.19 {+-} 0.08{sup o} (yaw). Spine region had little influence on overall targeting error, which was <1 mm for more than 95% of treatments (median, 0.48 mm). In the maximum motion case, target coverage decreased by 1.7% (from 92.1% to 90.4%) for the 20-Gy prescription isodose. Spinal cord volume receiving more than 8 Gy increased slightly, from 2.41 to 2.46 cm{sup 3}. Conclusions: Submillimeter targeting precision was obtained for fiducial-free spinal radiosurgery despite patient motion. Patient motion has little effect on the delivered dose distribution when image-guided correction of beam aiming is employed.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Resident Block-Structured Adaptive Mesh Refinement on Thousands of Graphics Processing Units, In: 2015 44th International Conference on Parallel Processing

Block-structured adaptive mesh refinement (AMR) is a technique that can be used when solving partial differential equations to reduce the number of cells necessary to achieve the required accuracy in areas of interest. These areas (shock fronts, material interfaces, etc.) are recursively covered with finer mesh patches that are grouped into a hierarchy of refinement levels. Despite the potential for large savings in computational requirements and memory usage without a corresponding reduction in accuracy, AMR adds overhead in managing the mesh hierarchy, adding complex communication and data movement requirements to a simulation. In this paper, we describe the design and implementation of a resident GPU-based AMR library, including: the classes used to manage data on a mesh patch, the routines used for transferring data between GPUs on different nodes, and the data-parallel operators developed to coarsen and refine mesh data. We validate the performance and accuracy of our implementation using three test problems and two architectures: an 8 node cluster, and 4,196 nodes of Oak Ridge National Laboratory's Titan supercomputer. Our GPU-based AMR hydrodynamics code performs up to 4.87× faster than the CPU-based implementation, and is scalable on 4,196 K20x GPUs using a combination of MPI and CUDA.

Computer Science↗

Design Principles for Effective Knowledge Discovery from Big Data

Big data phenomenon refers to the practice of collection and processing of very large data sets and associated systems and algorithms used to analyze these massive datasets. Architectures for big data usually range across multiple machines and clusters, and they commonly consist of multiple special purpose sub-systems. Coupled with the knowledge discovery process, big data movement offers many unique opportunities for organizations to benefit (with respect to new insights, business optimizations, etc.). However, due to the difficulty of analyzing such large datasets, big data presents unique systems engineering and architectural challenges. In this paper, we present three sys- tem design principles that can inform organizations on effective analytic and data collection processes, system organization, and data dissemination practices. The principles presented derive from our own research and development experiences with big data problems from various federal agencies, and we illustrate each principle with our own experiences and recommendations.

Architecture↗

Parallel Block Structured Adaptive Mesh Refinement on Graphics Processing Units

Block-structured adaptive mesh refinement is a technique that can be used when solving partial differential equations to reduce the number of zones necessary to achieve the required accuracy in areas of interest. These areas (shock fronts, material interfaces, etc.) are recursively covered with finer mesh patches that are grouped into a hierarchy of refinement levels. Despite the potential for large savings in computational requirements and memory usage without a corresponding reduction in accuracy, AMR adds overhead in managing the mesh hierarchy, adding complex communication and data movement requirements to a simulation. In this paper, we describe the design and implementation of a native GPU-based AMR library, including: the classes used to manage data on a mesh patch, the routines used for transferring data between GPUs on different nodes, and the data-parallel operators developed to coarsen and refine mesh data. We validate the performance and accuracy of our implementation using three test problems and two architectures: an eight-node cluster, and over four thousand nodes of Oak Ridge National Laboratory’s Titan supercomputer. Our GPU-based AMR hydrodynamics code performs up to 4.87× faster than the CPU-based implementation, and has been scaled to over four thousand GPUs using a combination of MPI and CUDA.

97 MATHEMATICS AND COMPUTING↗

A programmable shared-memory system for an array of processing-in-memory devices

Processing in memory (PIM), the concept of integrating processing directly with memory has been attracting a lot of attention, since PIM can assist in overcoming the throughput limitation caused by data movement between CPU and memory. The challenge, however, is that it requires the programmers to have a deep understanding of the PIM architecture to maximize the benefits such as data locality and parallel thread execution on multiple PIM devices. In this study, we present AnalyzeThat, a programmable shared-memory system for parallel data processing with PIM devices. Thematic to AnalyzeThat is a rich PIM-aware data structure (PADS), which is an encapsulation that integrally ties together the data, the analysis tasks and the runtime needed to interface with the PIM device array. The PADS abstraction provides (i) a sophisticated key-value data container that allows programmers to easily store data on multiple PIMs, (ii) a suite of parallel operations with which users can easily implement data analysis applications, and (iii) a runtime, hidden to programmers, which provides the mechanisms needed to overlay both the data and the tasks on the PIM device array in an intelligent fashion, based on PIM-specific information collected from the hardware. We have developed a PIM emulation framework called AnalyzeThat. In conclusion, our experimental evaluation with representative data analytics applications suggests that the proposed system can significantly reduce the PIM programming effort without losing its technology benefits.

97 MATHEMATICS AND COMPUTING↗

AnalyzeThat: A Programmable Shared-Memory System for an Array of Processing-In-Memory Devices

Processing In Memory (PIM), the concept of integrating processing directly with memory, has been attracting a lot of attention since PIM can assist in overcoming the throughput limitation caused by data movement between CPU and memory. The challenge, however, is that it requires the programmers to have a deep understanding of the PIM architecture to maximize the benefits such as data locality and parallel thread execution on multiple PIM devices. In this study, we present AnalyzeThat, a programmable shared-memory system for parallel data processing with PIM devices. Thematic to AnalyzeThat is a rich PIM-Aware Data Structure (PADS), which is an encapsulation that integrally ties together the data, the analysis tasks and the runtime needed to interface with the PIM device array. The PADS abstraction provides (i) a key-value data container that allows programmers to easily store data on multiple PIMs, (ii) a suite of parallel operations with which users can easily implement data analysis applications, and (iii) a runtime, hidden to programmers, which provides the mechanisms needed to overlay both the data and the tasks on the PIM device array in an intelligent fashion, based on PIM-specific information collected from the hardware. We have developed a PIM emulation framework called AnalyzeThat. Our experimental evaluation with representative data analytics applications suggests that the proposed system can significantly reduce the PIM programming effort without losing its technology benefits.

97 MATHEMATICS AND COMPUTING↗

Exploring power behaviors and trade-offs of in-situ data analytics, In: SC '13 Proceedings of the International Conference on High Performance Computing, Networking, Storage and Analysis

As scientific applications target exascale, challenges related to data and energy are becoming dominating concerns. For example, coupled simulation workflows are increasingly adopting in-situ data processing and analysis techniques to address costs and overheads due to data movement and I/O. However it is also critical to understand these overheads and associated trade-offs from an energy perspective. The goal of this paper is exploring data-related energy/performance trade-offs for end-to-end simulation workflows running at scale on current high-end computing systems. Specifically, this paper presents: (1) an analysis of the data-related behaviors of a combustion simulation workflow with an in-situ data analytics pipeline, running on the Titan system at ORNL; (2) a power model based on system power and data exchange patterns, which is empirically validated; and (3) the use of the model to characterize the energy behavior of the workflow and to explore energy/performance trade-offs on current as well as emerging systems.

Computer Science↗

Active Flash: Performance-Energy Tradeoffs for Out-of-Core Processing on Non-Volatile Memory Devices

In this abstract, we study the performance and energy tradeoffs involved in migrating data analysis into the flash device, a process we refer to as Active Flash. The Active Flash paradigm is similar to 'active disks', which has received considerable attention. Active Flash allows us to move processing closer to data, thereby minimizing data movement costs and reducing power consumption. It enables true out-of-core computation. The conventional definition of out-of-core solvers refers to an approach to process data that is too large to fit in the main memory and, consequently, requires access to disk. However, in Active Flash, processing outside the host CPU literally frees the core and achieves real 'out-of-core' analysis. Moving analysis to data has long been desirable, not just at this level, but at all levels of the system hierarchy. However, this requires a detailed study on the tradeoffs involved in achieving analysis turnaround under an acceptable energy envelope. To this end, we first need to evaluate if there is enough computing power on the flash device to warrant such an exploration. Flash processors require decent computing power to run the internal logic pertaining to the Flash Translation Layer (FTL), which is responsible for operations such as address translation, garbage collection (GC) and wear-leveling. Modern SSDs are composed of multiple packages and several flash chips within a package. The packages are connected using multiple I/O channels to offer high I/O bandwidth. SSD computing power is also expected to be high enough to exploit such inherent internal parallelism within the drive to increase the bandwidth and to handle fast I/O requests. More recently, SSD devices are being equipped with powerful processing units and are even embedded with multicore CPUs (e.g. ARM Cortex-A9 embedded processor is advertised to reach 2GHz frequency and deliver 5000 DMIPS; OCZ RevoDrive X2 SSD has 4 SandForce controllers, each with 780MHz max frequency Tensilica core). Efforts that take advantage of the available computing cycles on the processors on SSDs to run auxiliary tasks other than actual I/O requests are beginning to emerge. Kim et al. investigate database scan operations in the context of processing on the SSDs, and propose dedicated hardware logic to speed up scans. Also, cluster architectures have been explored, which consist of low-power embedded CPUs coupled with small local flash to achieve fast, parallel access to data. Processor utilization on SSD is highly dependent on workloads and, therefore, they can be idle during periods with no I/O accesses. We propose to use the available processing capability on the SSD to run tasks that can be offloaded from the host. This paper makes the following contributions: (1) We have investigated Active Flash and its potential to optimize the total energy cost, including power consumption on the host and the flash device; (2) We have developed analytical models to analyze the performance-energy tradeoffs for Active Flash, by treating the SSD as a blackbox, this is particularly valuable due to the proprietary nature of the SSD internal hardware; and (3) We have enhanced a well-known SSD simulator (from MSR) to implement 'on-the-fly' data compression using Active Flash. Our results provide a window into striking a balance between energy consumption and application performance.

42 ENGINEERING↗

The Geothermal Data Repository: Five Years of Open Geothermal Data, Benefits to the Community: Preprint

In the five years since its inception, the Department of Energy's (DOE) Geothermal Data Repository (GDR) has grown from the simple idea of storing public data in a centralized location to a valuable tool at the center of the DOE open data movement where it is providing a tangible benefit to the geothermal scientific community. Throughout this time, the GDR project team has been working closely with the community to refine the data submission process, improve the quality of submitted data, and embrace modern proper data management strategies to maximize the value and utility of submitted data. This paper explores some of the motivations behind various improvements to the GDR over the last 5 years, changes in data submission trends, and the ways in which these improvements have helped to drive research, fuel innovation, and accelerate the adoption of geothermal technologies.

15 GEOTHERMAL ENERGY↗

The Geothermal Data Repository: Five Years of Open Geothermal Data, Benefits to the Community

In the five years since its inception, the Department of Energy's (DOE) Geothermal Data Repository (GDR) has grown from the simple idea of storing public data in a centralized location to a valuable tool at the center of the DOE open data movement where it is providing a tangible benefit to the geothermal scientific community. Throughout this time, the GDR project team has been working closely with the community to refine the data submission process, improve the quality of submitted data, and embrace modern proper data management strategies to maximize the value and utility of submitted data. This paper explores some of the motivations behind various improvements to the GDR over the last 5 years, changes in data submission trends, and the ways in which these improvements have helped to drive research, fuel innovation, and accelerate the adoption of geothermal technologies.

15 GEOTHERMAL ENERGY↗

Data Locality Enhancement of Dynamic Simulations for Exascale Computing (Final Report)

The development of modern processors exhibits two trends that complicate the optimizations of modern software. The first is the increasing sensitivity of processors' throughput to irregularities in computation. With more processors produced through a massive integration of simple cores, future systems will increasingly favor regular data-level parallel computations, but deviate from the needs of applications with complex patterns. Some evidences are already shown on Graphic Processing Units (GPU): Irregular data accesses (e.g., indirect references A[D[i]]) and conditional branches are limiting many GPU applications' performance at a level an order of magnitude lower than the peak of GPU. The second hardware trend is the growing gap between memory bandwidth and the aggregate speed—that is, the sum of all cores' computing power—of a Chip Multiprocessor (CMP). Despite the capped growth of the peak CPU speed, the aggregate speed of a CMP keeps increasing as more cores get into a single chip. It is expected that by 2018, node concurrency in an exascale system will increase by hundreds of times, whereas, memory bandwidth will expand by only 10 to 20 times. Consequently, data movement and storage is expected to consume more than 70% of the total system power. Bridging this gap is difficult; the complexities of modern CMP memory hierarchy make it even harder: Data cache becomes shared among computing units, and the sharing is often non-uniform—whether two computing units share a cache depends on their proximity and the level of the cache. On the recent IBM Power7 architecture, for instance, four hardware contexts (or SMT threads) in a core share the entire memory hierarchy, all cores in one chip share an on-chip L3 cache, and cores across chips share L3 and main memory through off-chip connections. These two trends complicate the translation of computing power into performance, especially for a program with either intensive data accesses or complex patterns in data accesses or control flow paths. Unfortunately, both attributes present and will persist in a class of important applications. For instance, many scientific simulations deal with a large volume of data. And meanwhile, as most real-world processes are non-uniform and evolving (e.g., the evolution of a galaxy or the process of a drug injection), both the computations and data accesses of these programs tend to be irregular and dynamically changing. Currently, the lack of support to these applications on modern CMP severely limits their performance. On GPU, as our recent study shows and other studies echo, performance enhancement of a factor of integers is possible when memory accesses or control flows are streamlined for a set of GPU applications. On multicore CPU, our studies show that traditional locality enhancement, for being oblivious to the new features of multicore memory hierarchy, may even cause large slowdown to data-intensive dynamic applications. The severity of the issues is expected to worsen as the two hardware trends continue. Some recent studies try to match software with the trends, but in a limited scope or manner. For irregularities on GPU, most studies focus on irregularities analyzable through static analysis (e.g., data accesses in regular loops). Dynamic irregularities are harder to address because the needed analysis and transformations typically have to happen at run time. Some other research resorts to hardware extensions, an actual adoption of which is unclear for the entailed space cost and complexity. For data locality, recent years have seen some exploitations of the new memory hierarchy on multicore for performance, but most of them are on process or thread scheduling, rather than program transformations. Our recent study reveals that program-level transformations may magnify the scheduling benefits by a factor of seven, concluding that program-level transformation should play a central role for data locality enhancement on modern CMP. But research in this direction has been sparse, and most have focused on data layout or cache performance modeling, rather than program transformations to match with the new memory hierarchy features. Overall, it is still an open question how to bridge the gap between dynamic computations and the two prominent properties of modern processors. The goal of this project is to develop a set of techniques and software tools to enhance the matching between memory accesses in dynamic simulations and the prominent features of modern and future CMP, alleviating the memory performance issues for petascale and exascale computing. This report summarizes the discoveries and products produced throughout this project. It includes free launch, a new software approach to overcoming the shortcomings of both methods; coherence-free multiview, an approach that allows multiple views of a single data object to co-exist on GPU memory during a GPU kernel execution; algorithmic optimizations to data analytics problems, especially those that involve lots of distance calculations. Supporting dynamic parallelism is important for GPU to benefit a broad range of applications. There are currently two fundamental ways for programs to exploit dynamic parallelism on GPU: a software-based approach with software-managed worklists, and a hardware-based approach through dynamic subkernel launches. Neither is satisfactory. The former is complicated to program and is often subject to some load imbalance; the latter suffers large runtime overhead. In this work, we propose free launch, a new software approach to overcoming the shortcomings of both methods. It allows programmers to use subkernel launches to express dynamic parallelism. It employs a novel compiler-based code transformation named subkernel launch removal to replace the subkernel launches with the reuse of parent threads. Coupled with an adaptive task assignment mechanism, the transformation reassigns the tasks in the subkernels to the parent threads with a good load balance. The technique requires no hardware extensions, immediately deployable on existing GPUs. It keeps the programming convenience of the subkernel launch-based approach while avoiding its large runtime overhead. Meanwhile, its superior load balancing makes it outperform manual worklist-based techniques by 3X on average. The work was published at the 48th Annual IEEE/ACM International Symposium on Microarchitecture (Micro'2015). A Graphic Processing Unit (GPU) system is typically equipped with many types of memory (e.g., global, constant, texture, shared, cache). Data placement determines what data are placed on which type of memory, essential for GPU memory performance. Prior optimizations of data placement always require a single view of a data object on memory, which limits the optimization effectiveness. In this work, we propose coherence-free multiview, an approach that allows multiple views of a single data object to co-exist on GPU memory during a GPU kernel execution. We demonstrate that under certain conditions, the multiple views can remain incoherent while facilitating enhanced data placement. We present a theorem and some compiler support to ensure the soundness of the usage of coherence-free multiview. We further develop reference-discerning data placement, a new way to enhance data placements on GPU. It enables more flexible data placements by using coherence-free multiview to leverage the slack in coherence requirement of some GPU programs. Experiments on three types of GPU systems show that, with less than 200KB space cost, the new data placement technique can provide a 1.6X average (up to 4.27X) speedup. In addition, we have examined algorithmic optimizations to data analytics problems, especially those that involve lots of distance calculations. Computing distances among data points is an essential part of many important algorithms in data analytics, graph analysis, and other domains. In each of these domains, developers have spent significant manual effort optimizing algorithms, often through novel applications of the triangle equality, in order to minimize the number of distance computations in the algorithms. In this work, we observe that many algorithms across these domains can be generalized as an instance of a generic distance-related abstraction. Based on this abstraction, we derive seven principles for correctly applying the triangular inequality to optimize distance-related algorithms. Guided by the findings, we develop Triangular OPtimizer (TOP), the first software framework that is able to automatically produce optimized algorithms that either matches or outperforms manually designed algorithms for solving distance-related problems. TOP achieves up to 237x speedups and 2.5X on average. The work has been published at the 32nd International Conference on Machine Learning (ICML'15) and the 41st International Conference on Very Large Data Bases (VLDB'15).

42 ENGINEERING↗

Trends in data locality abstractions for HPC systems

The cost of data movement has always been an important concern in high performance computing (HPC) systems. It has now become the dominant factor in terms of both energy consumption and performance. Support for expression of data locality has been explored in the past, but those efforts have had only modest success in being adopted in HPC applications for various reasons. However, with the increasing complexity of the memory hierarchy and higher parallelism in emerging HPC systems, locality management has acquired a new urgency. Developers can no longer limit themselves to low-level solutions and ignore the potential for productivity and performance portability obtained by using locality abstractions. Fortunately, the trend emerging in recent literature on the topic alleviates many of the concerns that got in the way of their adoption by application developers. Data locality abstractions are available in the forms of libraries, data structures, languages and runtime systems; a common theme is increasing productivity without sacrificing performance. Furthermore, this paper examines these trends and identifies commonalities that can combine various locality concepts to develop a comprehensive approach to expressing and managing data locality on future large-scale high-performance computing systems.

97 MATHEMATICS AND COMPUTING↗