Search NASA⌕ Search

SEARCH · Search NASA

Results for “cache locality”

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 19 records
All sourcesNASA NTRSDOE OSTI
DOE OSTI2017

Cache Locality Optimization for Recursive Programs

We present an approach to optimize the cache locality for recursive programs by dynamically splicing--recursively interleaving--the execution of distinct function invocations. By utilizing data effect annotations, we identify concurrency and data reuse opportunities across function invocations and interleave them to reduce reuse distance. We present algorithms that efficiently track effects in recursive programs, detect interference and dependencies, and interleave execution of function invocations using user-level (non-kernel) lightweight threads. To enable multi-core execution, a program is parallelized using a nested fork/join programming model. Our cache optimization strategy is designed to work in the context of a random work stealing scheduler. We present an implementation using the MIT Cilk framework that demonstrates significant improvements in sequential and parallel performance, competitive with a state-of-the-art compile-time optimizer for loop programs and a domain- specific optimizer for stencil programs.

Lifflander, Jonathan↗
DOE OSTI2020

Resolution system that provides access to remote data locally

A system and method that provides access to remote data locally by identifying and downloading linked content from a plurality of remote sources through a communication network; decomposing the downloaded content into local content elements by intercepting links between the content elements and mapping the intercepted links to redirected links that locate the same download content in a local cache or a local server proxy; identifying remote devices requesting access to the local cache or the local server proxy and the device's functionality through a tracking code stored on the remote devices and a probabilistic tracking; and providing access to the local cache or the local server proxy in response to authorizations associated with the identified devices and a user.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2018

Light-weight cache coherence for data processors with limited data sharing

A data processing system includes a plurality of processors, local memories associated with a corresponding processor, and at least one inter-processor link. In response to a first processor performing a load or store operation on an address of a corresponding local memory that is not currently in the local cache, a local cache allocates a first cache line and encodes a local state with the first cache line. In response to a load operation from an address of a remote memory that is not currently in the local cache, the local cache allocates a second cache line and encodes a remote state with the second cache line. The first processor performs subsequent loads and stores on the first cache line in the local cache in response to the local state, and subsequent loads from the second cache line in the local cache in response to the remote state.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2015

Spatial Locality-Aware Cache Partitioning for Effective Cache Sharing, In: 2015 44th International Conference on Parallel Processing

In modern multi-core processors, last-level caches (LLCs) are typically shared among multiple cores. Previous works have shown that such sharing is beneficial as different workloads have different needs for cache capacity, and logical partitioning of capacity can improve system performance. However, what is missing in previous works on partitioning shared LLCs is that the heterogeneity in spatial locality among workloads has not been explored. In other words, all the cores use the same block/line size in shared LLCs. In this work, we highlight that exploiting spatial locality enables much more effective cache sharing. The fundamental reason is that for many memory intensive workloads, their cache capacity requirements can be drastically reduced when a large block size is employed, therefore they can effectively donate more capacity to other workloads. To leverage spatial locality for cache partitioning effectively, we first propose a simple yet effective mechanism to measure both spatial and temporal locality at run-time. The locality information is then used to determine both the proper block size and the capacity assigned to each workload. Our experiments show that our Spatial Locality-aware Cache Partitioning (SLCP) significantly outperforms the previous works. We also present several case studies that dissect the effectiveness of SLCP compared to the existing approaches.

Computer Science↗
DOE OSTI2013

Ordering of guarded and unguarded stores for no-sync I/O

A parallel computing system processes at least one store instruction. A first processor core issues a store instruction. A first queue, associated with the first processor core, stores the store instruction. A second queue, associated with a first local cache memory device of the first processor core, stores the store instruction. The first processor core updates first data in the first local cache memory device according to the store instruction. The third queue, associated with at least one shared cache memory device, stores the store instruction. The first processor core invalidates second data, associated with the store instruction, in the at least one shared cache memory. The first processor core invalidates third data, associated with the store instruction, in other local cache memory devices of other processor cores. The first processor core flushing only the first queue.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Software-Hardware Co-design of Heterogeneous SmartNIC System for Recommendation Models Inference and Training

Deep Learning Recommendation Models (DLRMs) are critical applications in various domains and have evolved as one of the single largest machine learning applications. Trillions of DLRM parameters exceed the on-chip memory capacity of GPUs. Large-scale multi-node systems are required for distributed DLRM inference and training, which suffer from the all-to-all communication bottleneck, mainly limiting the scalability of ever-growing DLRMs. In recent years, SmartNICs have evolved with coupled computation and communication capabilities providing opportunities for a powerful heterogeneous device in the system. However, there isn't such a distributed system that fully leverages the abundant smartNIC resources that resolve the scalability issue of DLRMs. In this work, we proposed a software-hardware co-design of a heterogeneous smartNIC system that resolves the communication bottleneck of distributed DLRMs, mitigates the memory bandwidth pressure, and improves computation efficiency. We provide a set of smartNIC designs of cache systems (including local cache and remote cache) and smartNIC computation kernels which reduce data movement, relieve memory lookup intensity, and improve the GPU's computation efficiency. In addition, we propose a graph algorithm that improves the data locality of queries within batches which optimizes the overall system performance with higher data reuse. Our evaluation shows that our system achieves 2.1x latency speedup for inference and 1.6x throughput speedup for training.

Guo, Anqi↗
DOE OSTI2024

Fault-Tolerant Deep Learning Cache with Hash Ring for Load Balancing in HPC Systems

Large-scale DL on HPC systems like Frontier and Summit uses distributed node-local caching to address scalability and performance challenges. However, as these systems grow more complex, the risk of node failures increases, and current caching approaches lack fault tolerance, jeopardizing large-scale training jobs. We analyzed six months of SLURM job logs from Frontier and found that over 30% of jobs failed after an average of 75 minutes. To address this, we propose fault-tolerance strategies that recache data lost from failed nodes using a hash ring technique for balanced data recaching in the distributed node-local caching, reducing reliance on the PFS. Our extensive evaluations on Frontier showed that the hash ring-based recaching approach reduced training time by approximately 25% compared to the approach that redirects I/O to the PFS after node failures and demonstrated effective load balancing of training data across nodes.

Lee, Seoyeong↗
DOE OSTI2016

Critical Points Based Register-Concurrency Autotuning for GPUs

The unprecedented prevalence of GPGPU is largely attributed to its abundant on-chip register resources, which allow massively concurrent threads and extremely fast context switch. However, due to internal memory capacity constraints, there is a tradeoff between the per-thread register usage and the overall concurrency. This becomes a design problem in terms of performance tuning, since the performance “sweet spot” which can be significantly affected by these two factors is generally unknown beforehand. In this paper, we propose an effective autotuning solution to quickly and efficiently select the optimal number of registers perthread for delivering the best GPU performance. Experiments on three generations of GPUs (Nvidia Fermi, Kepler and Maxwell) demonstrate that our simple strategy can achieve an average of 10% performance improvement while a max of 50% over the original version without modifying the user program. Additionally, to reduce local cache misses due to register spilling and further improve performance, we explore three optimization schemes (i.e. bypass L1 for global memory access, enlarge local L1 cache and spill into shared memory) and discuss their impact on performance on a Kepler GPU.

auto-tuning↗
DOE OSTI2022

SLIP (Surrogate Launching and Integration Platform)

SLIP (Surrogate Launching and Integration Platform) is a software ecosystem for downloading and running ML/AI benchmarks. SLIP automatically downloads and sets up the code and data running a benchmark. The large data sets and model code are cached locally with a specific ID within a designated cache directory.

Brown, Cade↗
DOE OSTI2019

Network-aware cache coherence protocol enhancement

A non-uniform memory access system includes several nodes that each have one or more processors, caches, local main memory, and a local bus that connects a node's processor(s) to its memory. The nodes are coupled to one another over a collection of point-to-point interconnects, thereby permitting processors in one node to access data stored in another node. Memory access time for remote memory takes longer than local memory because remote memory accesses have to travel across a communications network to arrive at the requesting processor. In some embodiments, inter-cache and main-memory-to-cache latencies are measured to determine whether it would be more efficient to satisfy memory access requests using cached copies stored in caches of owning nodes or from main memory of home nodes.

Roberts, David A.↗
DOE OSTI2018

Distributed caching for processing raw arrays

As applications continue to generate multi-dimensional data at exponentially increasing rates, fast analytics to extract meaningful results is becoming extremely important. The database community has developed array databases that alleviate this problem through a series of techniques. In-situ mechanisms provide direct access to raw data in the original format---without loading and partitioning. Parallel processing scales to the largest datasets. In-memory caching reduces latency when the same data are accessed across a workload of queries. However, we are not aware of any work on distributed caching of multi-dimensional raw arrays. In this paper, we introduce a distributed framework for cost-based caching of multi-dimensional arrays in native format. Given a set of files that contain portions of an array and an online query workload, the framework computes an effective caching plan in two stages. First, the plan identifies the cells to be cached locally from each of the input files by continuously refining an evolving R-tree index. In the second stage, an optimal assignment of cells to nodes that collocates dependent cells in order to minimize the overall data transfer is determined. We design cache eviction and placement heuristic algorithms that consider the historical query workload. A thorough experimental evaluation over two real datasets in three file formats confirms the superiority -- by as much as two orders of magnitude -- of the proposed framework over existing techniques in terms of cache overhead and workload execution time.

Zhao, Weijie↗
NASA NTRS2004

FPGA-Based, Self-Checking, Fault-Tolerant Computers

A proposed computer architecture would exploit the capabilities of commercially available field-programmable gate arrays (FPGAs) to enable computers to detect and recover from bit errors. The main purpose of the proposed architecture is to enable fault-tolerant computing in the presence of single-event upsets (SEUs). [An SEU is a spurious bit flip (also called a soft error) caused by a single impact of ionizing radiation.] The architecture would also enable recovery from some soft errors caused by electrical transients and, to some extent, from intermittent and permanent (hard) errors caused by aging of electronic components. A typical FPGA of the current generation contains one or more complete processor cores, memories, and highspeed serial input/output (I/O) channels, making it possible to shrink a board-level processor node to a single integrated-circuit chip. Custom, highly efficient microcontrollers, general-purpose computers, custom I/O processors, and signal processors can be rapidly and efficiently implemented by use of FPGAs. Unfortunately, FPGAs are susceptible to SEUs. Prior efforts to mitigate the effects of SEUs have yielded solutions that degrade performance of the system and require support from external hardware and software. In comparison with other fault-tolerant- computing architectures (e.g., triple modular redundancy), the proposed architecture could be implemented with less circuitry and lower power demand. Moreover, the fault-tolerant computing functions would require only minimal support from circuitry outside the central processing units (CPUs) of computers, would not require any software support, and would be largely transparent to software and to other computer hardware. There would be two types of modules: a self-checking processor module and a memory system (see figure). The self-checking processor module would be implemented on a single FPGA and would be capable of detecting its own internal errors. It would contain two CPUs executing identical programs in lock step, with comparison of their outputs to detect errors. It would also contain various cache local memory circuits, communication circuits, and configurable special-purpose processors that would use self-checking checkers. (The basic principle of the self-checking checker method is to utilize logic circuitry that generates error signals whenever there is an error in either the checker or the circuit being checked.) The memory system would comprise a main memory and a hardware-controlled check-pointing system (CPS) based on a buffer memory denoted the recovery cache. The main memory would contain random-access memory (RAM) chips and FPGAs that would, in addition to everything else, implement double-error-detecting and single-error-correcting memory functions to enable recovery from single-bit errors.

Some, Raphael↗
DOE OSTI2017

HPC formulations of optimization algorithms for tensor completion

Tensor completion is a powerful tool used to estimate or recover missing values in multi-way data. It has seen great success in domains such as product recommendation and healthcare. Tensor completion is most often accomplished via low-rank sparse tensor factorization, a computationally expensive non-convex optimization problem which has only recently been studied in the context of parallel computing. In this work, we study three optimization algorithms that have been successfully applied to tensor completion: alternating least squares (ALS), stochastic gradient descent (SGD), and coordinate descent (CCD++). We explore opportunities for parallelism on shared- and distributed-memory systems and address challenges such as memory- and operation-efficiency, load balance, cache locality, and communication. Among our advancements are a communication-efficient CCD++ algorithm, an ALS algorithm rich in level-3 BLAS routines, and an SGD algorithm which combines stratification with asynchronous communication. Furthermore, we show that introducing randomization during ALS and CCD++ can accelerate convergence. We evaluate our parallel formulations on a variety of real datasets on a modern supercomputer and demonstrate speedups through 16384 cores. Further, these improvements reduce time-to-solution from hours to seconds on real-world datasets. We show that after our optimizations, ALS is advantageous on parallel systems of small-to-moderate scale, while both ALS and CCD++ provide the lowest time-to-solution on large-scale distributed systems.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2012

Multi-core processing and scheduling performance in CMS

Commodity hardware is going many-core. We might soon not be able to satisfy the job memory needs per core in the current single-core processing model in High Energy Physics. In addition, an ever increasing number of independent and incoherent jobs running on the same physical hardware not sharing resources might significantly affect processing performance. It will be essential to effectively utilize the multi-core architecture. CMS has incorporated support for multi-core processing in the event processing framework and the workload management system. Multi-core processing jobs share common data in memory, such us the code libraries, detector geometry and conditions data, resulting in a much lower memory usage than standard single-core independent jobs. Exploiting this new processing model requires a new model in computing resource allocation, departing from the standard single-core allocation for a job. The experiment job management system needs to have control over a larger quantum of resource since multi-core aware jobs require the scheduling of multiples cores simultaneously. CMS is exploring the approach of using whole nodes as unit in the workload management system where all cores of a node are allocated to a multi-core job. Whole-node scheduling allows for optimization of the data/workflow management (e.g. I/O caching, local merging) but efficient utilization of all scheduled cores is challenging. Dedicated whole-node queues have been setup at all Tier-1 centers for exploring multi-core processing workflows in CMS. We present the evaluation of the performance scheduling and executing multi-core workflows in whole-node queues compared to the standard single-core processing workflows.

Hernandez, J. M.↗
DOE OSTI2021

becquerel (bq) v0.4.0

Becquerel is a Python package for analyzing nuclear spectroscopic measurements. The core functionalities are reading and writing different spectrum file types, fitting spectral features, rebinning spectrum counts to different bin edges, performing detector calibrations and interpreting measurement results. It also includes tools for visualizing radiation spectra and fits of different spectral features, as well as convenient access to tabulated nuclear data both from remote servers and local caches. It relies heavily on the standard scientific Python stack of numpy, scipy, matplotlib, pandas, and numba. It is intended to be general-purpose enough that it can be useful to anyone from an undergraduate taking a laboratory course to the advanced

Bandstra, Mark↗
DOE OSTI2025

becquerel (bq) v0.7.0

Becquerel is a Python package for analyzing nuclear spectroscopic measurements. The core functionalities are reading and writing different spectrum file types, fitting spectral features, rebinning spectrum counts to different bin edges, performing detector calibrations and interpreting measurement results. It also includes tools for visualizing radiation spectra and fits of different spectral features, as well as convenient access to tabulated nuclear data both from remote servers and local caches. It relies heavily on the standard scientific Python stack of numpy, scipy, matplotlib, pandas, and numba. It is intended to be general-purpose enough that it can be useful to anyone from an undergraduate taking a laboratory course to the advanced researcher.

Bandstra, Mark [Lawrence Berkeley National Laborat↗