Search NASA⌕ Search

SEARCH · Search NASA

Results for “Memory management”

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 91 records · Page 5

Crunching Large Graphs with Commodity Processors

Crunching large graphs is the basis of many emerging appli- cations, such as social network analysis and bioinformatics. Graph analytics algorithms exhibit little locality and therefore present significant performance challenges. Hardware multi- threading systems (e.g, Cray XMT) show that with enough concurrency, we can tolerate long latencies. Unfortunately, this solution is not available with commodity parts. Our goal is to develop a latency-tolerant system built out of commodity parts and mostly in software. The proposed system includes a runtime that supports a large number of lightweight contexts, full-bit synchronization and a memory manager that provides a high-latency but high-bandwidth global shared memory. This paper lays out the vision for our system, and justifies its feasibility with a performance analysis of the run- time for latency tolerance.

99 GENERAL AND MISCELLANEOUS↗

A self-defining hierarchical data system

The Self-Defining Data System (SDS) is a system which allows the creation of self-defining hierarchical data structures in a form which allows the data to be moved between different machine architectures. Because the structures are self-defining they can be used for communication between independent modules in a distributed system. Unlike disk-based hierarchical data systems such as Starlink's HDS, SDS works entirely in memory and is very fast. Data structures are created and manipulated as internal dynamic structures in memory managed by SDS itself. A structure may then be exported into a caller supplied memory buffer in a defined external format. This structure can be written as a file or sent as a message to another machine. It remains static in structure until it is reimported into SDS. SDS is written in portable C and has been run on a number of different machine architectures. Structures are portable between machines with SDS looking after conversion of byte order, floating point format, and alignment. A Fortran callable version is also available for some machines.

Bailey, J.↗

Four-Channel PC/104 MIL-STD-1553 Circuit Board

The mini bus interface card (miniBIC) is the first four-channel electronic circuit board that conforms to MIL-STD-1553 and to the electrical-footprint portion of PC/104. [MIL-STD-1553 is a military standard that encompasses a method of communication and electrical- interface requirements for digital electronic subsystems connected to a data bus. PC/104 is an industry standard for compact, stackable modules that are fully compatible (in architecture, hardware, and software) with personal-computer data- and power-bus circuitry.] Prior to the development of the miniBIC, only one- and two-channel PC/104 MIL-STD-1553 boards were available. To obtain four channels, it was necessary to include at least two boards in a PC/104 stack. In comparison with such a two-board stack, the miniBIC takes up less space, consumes less power, and is more reliable. In addition, the miniBIC includes 32 digital input/output channels. The miniBIC (see figure) contains four MIL-STD-1553B hybrid integrated circuits (ICs), four transformers, a field-programmable gate array (FPGA), and an Industry Standard Architecture (ISA) interface. Each hybrid IC includes a MILSTD-1553 dual transceiver, memory-management circuitry, processor interface logic circuitry, and 64Kx16 bits of shared static random access memory. The memory is used to configure message and data blocks. In addition, 23 16-bit registers are available for (1) configuring the hybrid IC for, and starting it in, various modes of operation; (2) reading the status of the functionality of the hybrid IC; and (3) resetting the hybrid IC to a known state. The miniBIC can operate as a remote terminal, bus controller, or bus monitor. The FPGA provides the chip-select and data-strobe signals needed for operation of the hybrid ICs. The FPGA also receives interruption signals and forwards them to the ISA bus. The ISA interface connects the address, data, and control interfaces of the hybrid ICs to the ISA backplane. Each channel is, in effect, a MIL-STD-1553 interface that can operate either independently of the others or else as a redundant version of one of the others. The transformer in each channel provides electrical isolation between the rest of the miniBIC circuitry and the bus to which that channel is connected.

Cox, Gary L.↗

C-MOS array design techniques: SUMC multiprocessor system study

The current capabilities of LSI techniques for speed and reliability, plus the possibilities of assembling large configurations of LSI logic and storage elements, have demanded the study of multiprocessors and multiprocessing techniques, problems, and potentialities. Evaluated are three previous systems studies for a space ultrareliable modular computer multiprocessing system, and a new multiprocessing system is proposed that is flexibly configured with up to four central processors, four 1/0 processors, and 16 main memory units, plus auxiliary memory and peripheral devices. This multiprocessor system features a multilevel interrupt, qualified S/360 compatibility for ground-based generation of programs, virtual memory management of a storage hierarchy through 1/0 processors, and multiport access to multiple and shared memory units.

Clapp, W. A.↗

Linear complexity

We present factorization and solution phases for a new linear complexity direct solver designed for concurrent batch operations on fine-grained parallel architectures, for matrices amenable to hierarchical representation. We focus on the strong-admissibility-based $\mathscr{H}^{2}$ format, where strong recursive skeletonization factorization compresses remote interactions. We build upon previous implementations of $\mathscr{H}^{2}$ matrix construction for efficient factorization and solution algorithm design, which are illustrated graphically in stepwise detail. The algorithms are ‘blackbox’ in the sense that the only inputs are the matrix and right-hand side, without analytical or geometrical information about the origin of the system. We demonstrate linear complexity scaling in both time and memory on four representative families of dense matrices up to one million in size. Parallel scaling up to 16 threads is enabled by a multi-level matrix graph coloring and avoidance of dynamic memory allocations thanks to prefix-sum memory management. An experimental backward error analysis is included. We break down the timings of different phases, identify phases that are memory-bandwidth limited, and discuss alternatives for phases that may be sensitive to the trend to employ lower precisions for performance.

Boukaram, Wajih↗

Formal verification of an MMU and MMU cache

We describe the formal verification of a hardware subsystem consisting of a memory management unit and a cache. These devices are verified independently and then shown to interact correctly when composed. The MMU authorizes memory requests and translates virtual addresses to real addresses. The cache improves performance by maintaining a LRU (least recently used) list from the memory resident segment table.

Schubert, E. T.↗

Tiling Framework for Heterogeneous Computing of Matrix based Tiled Algorithms

Tiling matrix operations can improve the load balancing and performance of applications on heterogeneous computing resources. Writing a tile-based algorithm for each operation with a traditional, hand-tuned tiling approach that uses for loops in C/C++ is cumbersome and error prone. Moreover, it must enable and support the heterogeneous memory management of data objects and also explore architecture-supported, native, tiled-data transfer APIs instead of copying the tiled data to continuous memory before the data transfer. The tiling framework provides a tiled data structure for heterogeneous memory mapping and parameterization to a heterogeneous task specification API. We have integrated our tiled framework into MatRIS (Math kernels library using IRIS). IRIS is a heterogeneous run-time framework with a heterogeneous programming model, memory model, and task execution model. Experiments reveal that the tiled framework for BLAS operations has improved the programmability of tiled BLAS and improved performance by ~20% when compared against the traditional method that copies the data to continuous memory locations for heterogeneous computing.

Miniskar, Narasinga Rao↗

Formal mechanization of device interactions with a process algebra

The principle emphasis is to develop a methodology to formally verify correct synchronization communication of devices in a composed hardware system. Previous system integration efforts have focused on vertical integration of one layer on top of another. This task examines 'horizontal' integration of peer devices. To formally reason about communication, we mechanize a process algebra in the Higher Order Logic (HOL) theorem proving system. Using this formalization we show how four types of device interactions can be represented and verified to behave as specified. The report also describes the specification of a system consisting of an AVM-1 microprocessor and a memory management unit which were verified in previous work. A proof of correct communication is presented, and the extensions to the system specification to add a direct memory device are discussed.

Schubert, E. Thomas↗

Out-of-Core Streamline Visualization on Large Unstructured Meshes

It's advantageous for computational scientists to have the capability to perform interactive visualization on their desktop workstations. For data on large unstructured meshes, this capability is not generally available. In particular, particle tracing on unstructured grids can result in a high percentage of non-contiguous memory accesses and therefore may perform very poorly with virtual memory paging schemes. The alternative of visualizing a lower resolution of the data degrades the original high-resolution calculations. This paper presents an out-of-core approach for interactive streamline construction on large unstructured tetrahedral meshes containing millions of elements. The out-of-core algorithm uses an octree to partition and restructure the raw data into subsets stored into disk files for fast data retrieval. A memory management policy tailored to the streamline calculations is used such that during the streamline construction only a very small amount of data are brought into the main memory on demand. By carefully scheduling computation and data fetching, the overhead of reading data from the disk is significantly reduced and good memory performance results. This out-of-core algorithm makes possible interactive streamline visualization of large unstructured-grid data sets on a single mid-range workstation with relatively low main-memory capacity: 5-20 megabytes. Our test results also show that this approach is much more efficient than relying on virtual memory and operating system's paging algorithms.

Ueng, Shyh-Kuang↗

Performance optimization, modeling and analysis of sparse matrix-matrix products on multi-core and many-core processors

Sparse matrix-matrix multiplication (SpGEMM) is a computational primitive that is widely used in areas ranging from traditional numerical applications to recent big data analysis and machine learning. Although many SpGEMM algorithms have been proposed, hardware specific optimizations for multi- and many-core processors are lacking and a detailed analysis of their performance under various use cases and matrices is not available. In this work, we firstly identify and mitigate multiple bottlenecks with memory management and thread scheduling on Intel Xeon Phi (Knights Landing or KNL). Specifically targeting many-core processors, we develop a hash-table-based algorithm and optimize a heap-based shared-memory SpGEMM algorithm. We examine their performance together with other publicly available codes. Different from the literature, our evaluation also includes use cases that are representative of real graph algorithms, such as multi-source breadth-first search or triangle counting. Our hash-table and heap-based algorithms are showing significant speedups from libraries in the majority of the cases while different algorithms dominate the other scenarios with different matrix size, sparsity, compression factor and operation type. We wrap up in-depth evaluation results and make a recipe to give the best SpGEMM algorithm for target scenario. We build the performance model for hash-table and heap-based algorithms, which supports the recipe. A critical finding is that hash-table-based SpGEMM gets a significant performance boost if the nonzeros are not required to be sorted within each row of the output matrix. Finally, we integrate our implementations into a large-scale protein clustering code named HipMCL, accelerating its SpGEMM kernel by up to 10X and achieving an overall performance boost for the whole HipMCL application by 2.6X.

97 MATHEMATICS AND COMPUTING↗

Advanced information processing system: Local system services

The Advanced Information Processing System (AIPS) is a multi-computer architecture composed of hardware and software building blocks that can be configured to meet a broad range of application requirements. The hardware building blocks are fault-tolerant, general-purpose computers, fault-and damage-tolerant networks (both computer and input/output), and interfaces between the networks and the computers. The software building blocks are the major software functions: local system services, input/output, system services, inter-computer system services, and the system manager. The foundation of the local system services is an operating system with the functions required for a traditional real-time multi-tasking computer, such as task scheduling, inter-task communication, memory management, interrupt handling, and time maintenance. Resting on this foundation are the redundancy management functions necessary in a redundant computer and the status reporting functions required for an operator interface. The functional requirements, functional design and detailed specifications for all the local system services are documented.

Burkhardt, Laura↗

High-Performance Sparse Matrix-Matrix Products on Intel KNL and Multicore Architectures

Sparse matrix-matrix multiplication (SpGEMM) is a computational primitive that is vastly used in areas ranging from traditional numerical applications to recent big data analysis and machine learning. While many SpGEMM algorithms have been proposed, hardware specific optimizations for multi- and many-core processors are lacking and a detailed analysis of their performance under various use cases and matrices is not available. We firstly identify and mitigate multiple bottlenecks with memory management and thread scheduling on Intel Xeon Phi (Knights Landing or KNL). Specifically targeting multi- and many-core processors, we develop a hash-table-based algorithm and optimize a heap-based shared-memory SpGEMM algorithm. We investigate their performance together with other publicly available codes. Different from the literature, our evaluation also includes use cases that are representative of real graph algorithms, such as multi-source breadth-first search or triangle counting. Our hash-table and heap-based algorithms are showing significant speedups from libraries in the majority of the cases while different algorithms dominate the other scenarios with different matrix size, sparsity, compression factor and operation type. We wrap up in-depth evaluation results and make a recipe to give the best SpGEMM algorithm for target scenario. In conclusion, a critical finding is that hash-table-based SpGEMM gets a significant performance boost if the nonzeros are not required to be sorted within each row of the output matrix.

97 MATHEMATICS AND COMPUTING↗

HAMR - Heterogeneous Accelerator Memory Resource (HAMR) v1.0

HAMR is a library defining an accelerator technology agnostic memory model that bridges between accelerator technologies (CUDA, HIP, ROCm, OpenMP, Sycl, OpenCL, Kokos, etc) and traditional CPUs in heterogeneous computing environments. HAMR is light weight and implemented in modern C++. HAMR can be used to manage memory with in a single code or as a data model for coupling codes in a technologically agnostic way. HAMR provides a Python module for coupling C++ and Python codes which implements zero-copy data transfers to and from Python using the Numpy array interface and Numba CUDA array interface protocols.

Loring, Burlen↗

A GPU-based compressible combustion solver for applications exhibiting disparate space and time scales

High-speed chemically active flows pose significant computational challenges due to their disparate space and time scales, with stiff chemistry often dominating simulation time. While modern scientific computing programs achieve exascale performance by leveraging graphics processing units (GPUs), existing GPU-based compressible combustion solvers face critical limitations in memory management, load balancing, and handling the highly localized nature of chemical reactions. To this end, we present a high-performance compressible reacting flow solver built on the AMReX framework and optimized for multi-GPU settings. Here, our approach addresses three GPU performance bottlenecks: memory access patterns through column-major storage optimization, computational workload variability via a bulk-sparse integration strategy for chemical kinetics, and multi-GPU load distribution for adaptive mesh refinement applications. The solver adapts existing matrix-based chemical kinetics formulations to multi-grid contexts. Using representative combustion applications, including 2D and 3D detonations and a 3D jet-in-crossflow configuration, we demonstrate 1.4–5× performance improvements over initial implementations on an in-house cluster of NVIDIA H100 GPUs, and near-ideal weak scaling on the Frontier supercomputer (Oak Ridge Leadership Computing Facility) with up to 1024 AMD Instinct MI250X GPUs. Roofline analysis reveals substantial improvements in arithmetic intensity for both convection (∼ 10 ×) and chemistry (∼ 4 ×) routines, confirming efficient utilization of GPU memory bandwidth and computational resources.

42 ENGINEERING↗

JavaGenes and Condor: Cycle-Scavenging Genetic Algorithms

A genetic algorithm code, JavaGenes, was written in Java and used to evolve pharmaceutical drug molecules and digital circuits. JavaGenes was run under the Condor cycle-scavenging batch system managing 100-170 desktop SGI workstations. Genetic algorithms mimic biological evolution by evolving solutions to problems using crossover and mutation. While most genetic algorithms evolve strings or trees, JavaGenes evolves graphs representing (currently) molecules and circuits. Java was chosen as the implementation language because the genetic algorithm requires random splitting and recombining of graphs, a complex data structure manipulation with ample opportunities for memory leaks, loose pointers, out-of-bound indices, and other hard to find bugs. Java garbage-collection memory management, lack of pointer arithmetic, and array-bounds index checking prevents these bugs from occurring, substantially reducing development time. While a run-time performance penalty must be paid, the only unacceptable performance we encountered was using standard Java serialization to checkpoint and restart the code. This was fixed by a two-day implementation of custom checkpointing. JavaGenes is minimally integrated with Condor; in other words, JavaGenes must do its own checkpointing and I/O redirection. A prototype Java-aware version of Condor was developed using standard Java serialization for checkpointing. For the prototype to be useful, standard Java serialization must be significantly optimized. JavaGenes is approximately 8700 lines of code and a few thousand JavaGenes jobs have been run. Most jobs ran for a few days. Results include proof that genetic algorithms can evolve directed and undirected graphs, development of a novel crossover operator for graphs, a paper in the journal Nanotechnology, and another paper in preparation.

Globus, Al↗

Architecture independent environment for developing engineering software on MIMD computers

Engineers are constantly faced with solving problems of increasing complexity and detail. Multiple Instruction stream Multiple Data stream (MIMD) computers have been developed to overcome the performance limitations of serial computers. The hardware architectures of MIMD computers vary considerably and are much more sophisticated than serial computers. Developing large scale software for a variety of MIMD computers is difficult and expensive. There is a need to provide tools that facilitate programming these machines. First, the issues that must be considered to develop those tools are examined. The two main areas of concern were architecture independence and data management. Architecture independent software facilitates software portability and improves the longevity and utility of the software product. It provides some form of insurance for the investment of time and effort that goes into developing the software. The management of data is a crucial aspect of solving large engineering problems. It must be considered in light of the new hardware organizations that are available. Second, the functional design and implementation of a software environment that facilitates developing architecture independent software for large engineering applications are described. The topics of discussion include: a description of the model that supports the development of architecture independent software; identifying and exploiting concurrency within the application program; data coherence; engineering data base and memory management.

Valimohamed, Karim A.↗

Metall: A persistent memory allocator for data-centric analytics

Data analytics applications transform raw input data into analytics-specific data structures before performing analytics. Unfortunately, such data ingestion steps are often more expensive than analytics. In addition, various types of NVRAM devices are already used in many HPC systems today. Such devices will be useful for storing and reusing data structures beyond a single process life cycle. We developed Metall, a persistent memory allocator built on top of the memory-mapped file mechanism. Metall enables applications to transparently allocate custom C++ data structures into various types of persistent memories. Metall incorporates a concise and high-performance memory management algorithm inspired by Supermalloc and the rich C++ interface developed by Boost.Interprocess library. On a dynamic graph construction workload, Metall achieved up to 11.7x and 48.3x performance improvements over Boost.Interprocess and memkind (PMEM kind), respectively. We also demonstrate Metall’s high adaptability by integrating Metall into a graph processing framework, GraphBLAS Template Library. Here this study’s outcomes indicate that Metall will be a strong tool for accelerating future large-scale data analytics by allowing applications to leverage persistent memory efficiently.

97 MATHEMATICS AND COMPUTING↗

Application-Controlled Demand Paging for Out-of-Core Visualization

In the area of scientific visualization, input data sets are often very large. In visualization of Computational Fluid Dynamics (CFD) in particular, input data sets today can surpass 100 Gbytes, and are expected to scale with the ability of supercomputers to generate them. Some visualization tools already partition large data sets into segments, and load appropriate segments as they are needed. However, this does not remove the problem for two reasons: 1) there are data sets for which even the individual segments are too large for the largest graphics workstations, 2) many practitioners do not have access to workstations with the memory capacity required to load even a segment, especially since the state-of-the-art visualization tools tend to be developed by researchers with much more powerful machines. When the size of the data that must be accessed is larger than the size of memory, some form of virtual memory is simply required. This may be by segmentation, paging, or by paged segments. In this paper we demonstrate that complete reliance on operating system virtual memory for out-of-core visualization leads to poor performance. We then describe a paged segment system that we have implemented, and explore the principles of memory management that can be employed by the application for out-of-core visualization. We show that application control over some of these can significantly improve performance. We show that sparse traversal can be exploited by loading only those data actually required. We show also that application control over data loading can be exploited by 1) loading data from alternative storage format (in particular 3-dimensional data stored in sub-cubes), 2) controlling the page size. Both of these techniques effectively reduce the total memory required by visualization at run-time. We also describe experiments we have done on remote out-of-core visualization (when pages are read by demand from remote disk) whose results are promising.

Cox, Michael↗