Search NASA⌕ Search

SEARCH · Search NASA

Results for “Memory systems”

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 73 records · Page 4

Runtime extension for neural network training with heterogeneous memory

Systems, apparatuses, and methods for managing buffers in a neural network implementation with heterogeneous memory are disclosed. A system includes a neural network coupled to a first memory and a second memory. The first memory is a relatively low-capacity, high-bandwidth memory while the second memory is a relatively high-capacity, low-bandwidth memory. During a forward propagation pass of the neural network, a run-time manager monitors the usage of the buffers for the various layers of the neural network. During a backward propagation pass of the neural network, the run-time manager determines how to move the buffers between the first and second memories based on the monitored buffer usage during the forward propagation pass. As a result, the run-time manager is able to reduce memory access latency for the layers of the neural network during the backward propagation pass.

Mappouras, Georgios↗

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↗

Memory access system

A method and system for accessing a memory for a data processing system. The method comprises sending a read request for a plurality of locations in the memory to read the plurality of locations in parallel based on an upper bound for reading the memory. The upper bound for a number of locations is based on a group of constraints for the memory. The method receives a summed value of a plurality of memory values in the plurality of locations in the memory.

97 MATHEMATICS AND COMPUTING↗

Stateless and stateful implementations of faithful execution

A faithful execution system includes system memory, a target processor, and protection engine. The system memory stores a ciphertext including value fields and integrity fields. The value fields each include an encrypted executable instruction and the integrity fields each include an encrypted integrity value for determining whether a corresponding one of the value fields has been modified. The target processor executes plaintext instructions decoded from the ciphertext while the protection engine is coupled between the system memory and the target processor. The protection engine includes logic to retrieve the ciphertext from the system memory, decrypt the value fields into the plaintext instructions, perform an integrity check based on the integrity fields to determine whether any of the corresponding value fields have been modified, and provide the plaintext instructions to the target processor for execution.

97 MATHEMATICS AND COMPUTING↗

Evaluating the effectiveness of program data features for guiding memory management

Recent trends have led to the adoption of larger and more complex memory systems, often with multiple tiers of memory performance within the same platform. To utilize complex memory systems efficiently, current data management strategies must be altered to map usage demands to the underlying hardware. Applications, as the generators of memory accesses, are well-suited to guide this process, but building and maintaining separate source code versions for different memory systems is not feasible in most cases. One potential solution is to employ automated program profiling and analysis to facilitate the production of application-based guidance. By attaching memory usage information to static or lightweight program features, compilers and runtime systems can generate fine-grained guidance without additional efforts from users or developers. Recent works have employed this approach with some success, but it is not clear which program features are most useful for guiding data management.This work evaluates the effectiveness of using different program data features to predict memory usage and guide memory management. It employs a custom set of simulation tools, based in the Intel® Pin framework, to collect and analyze the usage characteristics of application data associated with common program data features, such as allocation sites, types, and context. The results show that even relatively simple features, such as object size, are useful for selecting data with similar usage properties, but finer-grained features, such as the instructions that access a particular object, are often much more effective. Additionally, this work evaluates the performance of using different data features and different program inputs to guide data placement on a heterogeneous memory platform with a limited amount of high performance memory.

Effler, T. Chad↗

Compiler-Assisted Detection of Transient Memory Errors

The probability of bit flips in hardware memory systems is projected to increase significantly as memory systems continue to scale in size and complexity. Effective hardware-based error detection and correction requires that the complete data path, involving all parts of the memory system, be protected with sufficient redundancy. First, this may be costly to employ on commodity computing platforms and second, even on high-end systems, protection against multi-bit errors may be lacking. Therefore, augmenting hardware error detection schemes with software techniques is of consider- able interest. In this paper, we consider software-level mechanisms to comprehensively detect transient memory faults. We develop novel compile-time algorithms to instrument application programs with checksum computation codes so as to detect memory errors. Unlike prior approaches that employ checksums on computational and architectural state, our scheme verifies every data access and works by tracking variables as they are produced and consumed. Experimental evaluation demonstrates that the proposed comprehensive error detection solution is viable as a completely software-only scheme. We also demonstrate that with limited hardware support, overheads of error detection can be further reduced.

def-use analysis↗

Data Movement Dominates: Final Report

Over the past three years in this project, what we have observed is that the primary reason for data movement in large-scale systems is that the per-node capacity is not large enough—i.e., one of the solutions to the data-movement problem (certainly not the only solution that is required, but a significant one nonetheless) is to increase per-node capacity so that inter-node traffic is reduced. This unfortunately is not as simple as it sounds. Today’s main memory systems for datacenters, enterprise computing systems, and supercomputers, fail to provide high per-socket capacity [Dirik & Jacob 2009; Cooper-Balis et al. 2012], except at extremely high price points (factors of 10–100x the cost/bit of consumer main-memory systems) [Stokes 2008]. The reason is that our choice of technology for today’s main memory systems—i.e., DRAM, which we have used as a main-memory technology since the 1970s [Jacob et al. 2007]—can no longer keep up with our needs for density and price per bit. Main memory systems have always been built from the cheapest, densest, lowest-power memory technology available, and DRAM is no longer the cheapest, the densest, nor the lowest-power storage technology out there. It is now time for DRAM to go the way that SRAM went: move out of the way for a cheaper, slower, denser storage technology, and become a cache instead. This inflection point has happened before, in the context of SRAM yielding to DRAM. There was once a time that SRAM was the storage technology of choice for all main memories [Tomasulo 1967; Thornton 1970; Kidder 1981]. However, once DRAM hit volume production in the 1970s and 80s, it supplanted SRAM as a main memory technology because it was cheaper, and it was denser. It also happened to be lower power, but that was not the primary consideration of the day. At the time, it was recognized that DRAM was much slower than SRAM, but it was only at the supercomputer level (For instance the Cray X-MP in the 1980s and its follow-on, the Cray Y-MP, in the 1990s) that could one afford to build ever- larger main memories out of SRAM—the reasoning for moving to DRAM was that an appropriately designed memory hierarchy, built of DRAM as main memory and SRAM as a cache, would approach the performance of SRAM, at the price-per-bit of DRAM [Mashey 1999]. Today it is quite clear that, were one to build an entire multi-gigabyte main memory out of SRAM instead of DRAM, one could improve the performance of almost any computer system by up to an order of magnitude—but this option is not even considered, because to build that system would be prohibitively expensive. It is now time to revisit the same design choice in the context of modern technologies and modern systems. For reasons both technical and economic, we can no longer afford to build ever-larger main memory systems out of DRAM. Flash memory, on the other hand, is significantly cheaper and denser than DRAM and therefore should take its place. While it is true that flash is significantly slower than DRAM, one can afford to build much larger main memories out of flash than out of DRAM, and we show that an appropriately designed memory hierarchy, built of flash as main memory and DRAM as a cache, will approach the performance of DRAM, at the price-per-bit of flash. In our studies as part of this project, we have investigated Non-Volatile Main Memory (NVMM), a new main-memory architecture for large-scale computing systems, one that is specifically designed to address the weaknesses described previously. In particular, it provides the following features: non-volatility: The bulk of the storage is comprised of NAND flash, and in this organization DRAM is used only as a cache, not as main memory. Furthermore, the flash is journaled, which means that operations such as checkpoint/restore are already built into the system. 1+ terabytes of storage per socket: SSDs and DRAM DIMMs have roughly the same form factor (several square inches of PCB surface area), and terabyte SSDs are now commonplace. performance approaching that of DRAM: DRAM is used as a cache to the flash system. price-per-bit approaching that of NAND: Flash is currently well under $0.50 per gigabyte; DDR3 SDRAM is currently just over $10 per gigabyte [Newegg 2014]. Even today, one can build an easily affordable main memory system with a terabyte or more of NAND storage per CPU socket (which would be extremely expensive were one to use DRAM), and our cycle- accurate, full-system experiments show that this can be done at a performance point that lies within a factor of two of DRAM.

96 KNOWLEDGE MANAGEMENT AND PRESERVATION↗

Method and apparatus for scheduling in a non-uniform compute device

A data processing apparatus, and method of operation thereof, for executing instructions. The apparatus includes one or more host processors, each having a first processing unit, and a multi-level memory system. One or more levels of the memory system are tightly coupled to a corresponding second processing unit. At least one of the host processors includes an instruction scheduler that routes instructions selectively to at least one of the first and second processing units, dependent upon the availability of the processing units and the location, within the memory system, of data to be used when executing the instructions.

97 MATHEMATICS AND COMPUTING↗

Really Embedding Domain-Specific Languages into C++

The following topics are dealt with: program compilers; optimising compilers; parallel processing; software engineering; learning (artificial intelligence); multiprocessing systems; shared memory systems; optimisation; computational complexity; specification languages.

Finkel, Hal J.↗

Static power reduction for midpoint-terminated busses

A memory system is disclosed which is comprised of a memory controller and addressable memory devices such as DRAMs. The invention provides a programmable register to control the high vs. low drive state of each bit of a memory system address and control bus during periods of bus inactivity. In this way, termination voltage supply current can be minimized, while permitting selected bus bits to be driven to a required state. This minimizes termination power dissipation while not affecting memory system performance. The technique can be extended to work for other high-speed busses as well.

97 MATHEMATICS AND COMPUTING↗

A Processable Shape Memory Polymer System for Biomedical Applications

Polyurethane shape memory polymers (SMPs) with tunable thermomechanical properties and advanced processing capabilities are synthesized, characterized, and implemented in the design of a microactuator medical device prototype. The ability to manipulate glass transition temperature ( T g ) and crosslink density in low-molecular weight aliphatic thermoplastic polyurethane SMPs is demonstrated using a synthetic approach that employs UV catalyzed thiol-ene “click” reactions to achieve postpolymerization crosslinking. Polyurethanes containing varying C=C functionalization are synthesized, solution blended with polythiol crosslinking agents and photoinitiator and subjected to UV irradiation, and the effects of number of synthetic parameters on crosslink density are reported. Thermomechanical properties are highly tunable, including glass transitions tailorable between 30 and 105 °C and rubbery moduli tailorable between 0.4 and 20 MPa. This new SMP system exhibits high toughness for many formulations, especially in the case of low crosslink density materials, for which toughness exceeds 90 MJ m –3 at select straining temperatures. To demonstrate the advanced processing capability and synthetic versatility of this new SMP system, a laser-actuated SMP microgripper device for minimally invasive delivery of endovascular devices is fabricated, shown to exhibit an average gripping force of 1.43 ± 0.37 N and successfully deployed in an in vitro experimental setup under simulated physiological conditions.

42 ENGINEERING↗

Machine learning based design space exploration for hybrid main-memory design

We develop a machine learning (ML) based design space exploration (DSE) method that builds predictive models for various responses of a hybrid main-memory system. To overcome the challenges associated with latency, capacity, and power of memory systems in future extreme-scale machines, the hybrid memory architectures are being considered in which novel non-volatile memory (NVM) systems augment the traditional DRAM. However, way before their actual design and implementation, these emerging hybrid memory systems need to be simulated and analyzed to fully understand their capabilities and limitations. As the conventional architectural-level memory simulators require significant amounts of computational costs and time, we propose to utilize ML techniques for developing various memory-response models that can instantly provide a predicted response corresponding to any new memory configuration. Specifically, in this work, we apply four supervised ML techniques to build regression models for memory latency, bandwidth, power, and total read/write responses. The training and validation data for the ML methods are generated using NVMain memory simulator for DRAM, NVM, and their hybrid combinations. We demonstrate the results of the ML based memory-DSE method in terms of the learning curve characteristics for hyperparameter tuning and the statistical error analyses of the designed predictive models.

Sen, Satyabrata↗

Method and apparatus for reordering in a non-uniform compute device

A data processing apparatus includes a multi-level memory system, one or more first processing unit coupled to the memory system at a first level and one or more second processing units each coupled to the memory system at a second level. A first reorder buffer maintains data order during execution of instructions by the first and second processing units and a second reorder buffer maintains data order during execution of the instructions by an associated second processing unit. An entry in the first reorder buffer is configured, dependent upon an indicator bit, as an entry for a single instruction or a pointer to an entry in the second reorder buffer. An entry in the second reorder buffer includes instruction block start and end addresses and indicators of input and output register. Instructions are released to a processing unit when all inputs, as indicated by the reorder buffers, are available.

97 MATHEMATICS AND COMPUTING↗

Selective data retrieval based on access latency

A processor includes multiple processing units (e.g., processor cores), with each processing unit associated with at least one private, dedicated cache. The processor is also associated with a system memory that stores all data that can be accessed by the multiple processing units. A coherency manager (e.g., a coherence directory) of the processor enforces a specified coherency scheme to ensure data coherency between the different caches and between the caches and the system memory. In response to a memory access request to a given cache resulting in a cache miss, the coherency manager identifies the current access latency to the system memory as well as the current access latencies to other caches of the processor. The coherency manager transfers the targeted data to the given cache from the cache or system memory having the lower access latency.

97 MATHEMATICS AND COMPUTING↗

Tau deposition is associated with functional isolation of the hippocampus in aging

The tau protein aggregates in aging and Alzheimer disease and may lead to memory loss through disruption of medial temporal lobe (MTL)-dependent memory systems. Here, we investigated tau-mediated mechanisms of hippocampal dysfunction that underlie the expression of episodic memory decline using fMRI measures of hippocampal local coherence (regional homogeneity; ReHo), distant functional connectivity and tau-PET. We show that age and tau pathology are related to higher hippocampal ReHo. Functional disconnection between the hippocampus and other components of the MTL memory system, particularly an anterior-temporal network specialized for object memory, is also associated with higher hippocampal ReHo and greater tau burden in anterior-temporal regions. These associations are not observed in the posteromedial network, specialized for context/spatial information. Higher hippocampal ReHo predicts worse memory performance. These findings suggest that tau pathology plays a role in disconnecting the hippocampus from specific MTL memory systems leading to increased local coherence and memory decline.

59 BASIC BIOLOGICAL SCIENCES↗

Design Space Exploration of Emerging Memory Technologies for Machine Learning Applications

Memory design space exploration methods study memory systems’ performances and limitations before implementation. The computer memory design space has grown exponentially because of the enormous growth of memory types, memory controllers, and application software. Computer simulators are commonly used for memory design space exploration. However, complex memory simulations take an enormous amount of time. Hence, in this paper, we proposed a machine learning-based design space exploration method for dynamic random-access memory and non-volatile memory systems. We applied our method to the CosmoGAN and LeNet applications to predict the following six memory response parameters: (i) bandwidth, (ii) power, (iii) average latency, (iv) average total latency, (v) memory reads, and (vi) memory writes. Our experimental results show that machine learning models can predict memory response parameter values faster than simulations. We used support vector machine, random forest, and gradient boosting machine learning models. We observed that the support vector machine provides better performance for bandwidth, average latency, and average total latency. The random forest model works better for memory reads and writes. The gradient boosting model provides superior prediction performance for power. We provide a detailed discussion on learning curve characteristics, error analysis, and memory type recommendation.

Hasan, S M Shamimul↗

Method and apparatus for fast context cloning in a data processing system

A data processing system includes a memory system, a first processing element, a first address translator that maps virtual addresses to system addresses, a second address translator that maps system address to physical addresses, and a task management unit. A first program task uses a first virtual memory space that is mapped to a first system address range using a first table. The context of the first program task includes an address of the first table and is cloned by creating a second table indicative of a mapping from a second virtual address space to a second range of system addresses, where the second range is mapped to the same physical addresses as the first range until a write occurs, at which time memory is allocated and the mapping of the second range is updated. The cloned context includes an address of the second table.

Beard, Jonathan Curtis↗

Flexible optimized data handling in systems with multiple memories

Methods and systems for optimizing an application for a computing system having multiple distinct memory locations that are interconnected by one or more communication channels include determining one or more data handling properties for a data region in an application. One or more data handling policies for the data region are determined based on the one or more data handling properties. Data setup costs are determined for a scope in the application that uses the data region in different memory locations based on the one or more data handling properties. The application is optimized in accordance with the one or more data handling policies and the data setup costs for the different memory locations.

Chen, Tong↗