Engineering Papers⌕ Search

SEARCH · Engineering Papers

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 73 records · Page 4

Extreme-bandwidth scalable performance-per-watt GPU architecture

A technique for accessing memory in an accelerated processing device coupled to stacked memory dies is provided herein. The technique includes receiving a memory access request from an execution unit and identifying whether the memory access request corresponds to memory cells of the stacked dies that are considered local to the execution unit or non-local. For local accesses, the access is made “directly”, that is, without using a bus. A control die coordinates operations for such local accesses, activating particular through-silicon-vias associated with the memory cells that include the data for the access. Non-local accesses are made via a distributed cache fabric and an interconnect bus in the control die. Various other features and details are provided below.

97 MATHEMATICS AND COMPUTING↗

Deep Reinforcement Learning-Based Edge Caching in Wireless Networks

With the purpose to offload data traffic in wireless networks, content caching techniques have recently been studied intensively. Using these techniques and caching a portion of the popular files at the local content servers, the users can be served with less delay. Most of the content replacement policies are based on the content popularity, that depends on the users' preferences. In practice, such information varies over time. Therefore, an approach to determine the file popularity patterns must be incorporated into caching policies. In this context, we study content caching at the wireless network edge using a deep reinforcement learning framework with Wolpertinger architecture. In particular, we propose deep actor-critic reinforcement learning based policies for both centralized and decentralized content caching. For centralized edge caching, we aim at maximizing the cache hit rate. In decentralized edge caching, we consider both the cache hit rate and transmission delay as performance metrics. These proposed frameworks are assumed to neither have any prior information on the file popularities nor know the potential variations in such information. Via simulation results, the superiority of the proposed frameworks is verified by comparing them with other policies, including least frequently used (LFU), least recently used (LRU), and first-in-first-out (FIFO) policies.

42 ENGINEERING↗

XRootd, disk-based, caching proxy for optimization of data access, data placement and data replication

Following the success of the XRootd-based US CMS data federation, the AAA project investigated extensions of the federation architecture by developing two sample implementations of an XRootd, disk-based, caching proxy. The first one simply starts fetching a whole file as soon as a file open request is received and is suitable when completely random file access is expected or it is already known that a whole file be read. The second implementation supports on-demand downloading of partial files. Extensions to the Hadoop Distributed File System have been developed to allow for an immediate fallback to network access when local HDFS storage fails to provide the requested block. Both cache implementations are in pre-production testing at UCSD.

Bauerdick, L. A.T.↗

GraphTango: A Hybrid Representation Format for Efficient Streaming Graph Updates and Analysis

Abstract Streaming graph processing performs batched updates and analytics on a time-evolving graph. The underlying representation format of the graph largely determines the throughputs of these updates and analytics phases. Existing representation formats usually employ variations of hash tables or adjacency lists. However, a recent study showed that the adjacency-list-based approaches perform poorly on heavy-tailed graphs, and the hash table-based approaches suffer on short-tailed graphs. We propose GraphTango, a hybrid representation format that provides excellent update and analytics throughput regardless of the graph’s degree distribution. GraphTango dynamically switches among three different formats based on a vertex’s degree: (i) Low-degree vertices store the edges directly with the neighborhood metadata, confining accesses to a single cache line, (2) Medium-degree vertices use adjacency lists, and (3) High-degree vertices use hash tables as well as adjacency lists. In this case, the adjacency list provides fast traversal during the analytics phase, while the hash table provides constant-time lookups during the update phase. We further optimized the performance by designing an open-addressing-based hash table that fully utilizes every fetched cache line. In addition, we developed a thread-local lock-free memory pool that allows fast growing/shrinking of the adjacency lists and hash tables in a multi-threaded environment. We evaluated GraphTango with the help of the SAGA-Bench framework and compared it with four other representation formats: Stinger, Degree-aware Robin Hood Hashing, and two adjacency list-based formats with different workload balancing scheme. On average, GraphTango provides 4.5x higher insertion throughput, 3.2x higher deletion throughput, and 1.1x higher analytics throughput over the next best format. Furthermore, we integrated GraphTango with the state-of-the-art graph processing frameworks DZiG and RisGraph. Compared to the vanilla DZiG and vanilla RisGraph , [ GraphTango + DZiG ] and [ GraphTango + RisGraph ] reduces the average batch processing time by 2.3x and 1.5x, respectively.

Ahmed, Alif↗

DDStore: Distributed Data Store for Scalable Training of Graph Neural Networks on Large Atomistic Modeling Datasets

Graph neural networks (GNNs) are a class of Deep Learning models used in designing atomistic materials for effective screening of large chemical spaces. To ensure robust prediction, GNN models must be trained on large volumes of atomistic data on leadership class supercomputers. Even with the advent of modern architectures that consist of multiple storage layers that include node-local NVMe devices in addition to device memory for caching large datasets, extreme-scale model training faces I/O challenges at scale.We present DDStore, an in-memory distributed data store designed for GNN training on large-scale graph data. DDStore provides a hierarchical, distributed, data caching technique that combines data chunking, replication, low-latency random access, and high throughput communication. DDStore achieves near-linear scaling for training a GNN model using up to 1000 GPUs on the Summit and Perlmutter supercomputers, and reaches up to a 6.15x reduction in GNN training time compared to state-of-the-art methodologies.

Choi, Jong Youl↗

Hvac: Removing I/O Bottleneck for Large-Scale Deep Learning Applications

Scientific communities are increasingly adopting deep learning (DL) models in their applications to accelerate scientific discovery processes. However, with rapid growth in the computing capabilities of HPC supercomputers, large-scale DL applications have to spend a significant portion of training time performing I/O to a parallel storage system. Previous research works have investigated optimization techniques such as prefetching and caching. Unfortunately, there exist non-trivial challenges to adopting the existing solutions on HPC supercomputers for large-scale DL training applications, which include non-performance and/or failures at extreme scale, lack of portability and generality in design, complex deployment methodology, and being limited to a specific application or dataset. To address these challenges, we propose High-Velocity AI Cache (HVAC), a distributed read-cache layer that targets and fully exploits the node-local storage or near node-local storage technology. HVAC seamlessly accelerates read I/O by aggregating node-local or near node-local storage, avoiding metadata lookups and file locking while preserving portability in the application code. We deploy and evaluate HVAC on 1,024 nodes (with over 6000 NVIDIA V100 GPUS) of the Summit supercomputer. In particular, we evaluate the scalability, efficiency, accuracy, and load distribution of HVAC compared to GPFS and XFS-on-NVMe. With four different DL applications, we observe an average 25 % performance improvement atop GPFS and 9% drop against XFS-on-NVMe, which scale linearly and are considered the performance upper bound. We envision HVAC as an important caching library for upcoming HPC supercomputers such as Frontier.

Khan, Awais↗

Optimizing Excited-State Electronic-Structure Codes for Intel Knights Landing: A Case Study on the BerkeleyGW Software

We profile and optimize calculations performed with the BerkeleyGW code on the Xeon-Phi architecture. BerkeleyGW depends both on hand-tuned critical kernels as well as on BLAS and FFT libraries. We describe the optimization process and performance improvements achieved. We discuss a layered parallelization strategy to take advantage of vector, thread and node-level parallelism. We discuss locality changes (including the consequence of the lack of L3 cache) and effective use of the on-package high-bandwidth memory. We show preliminary results on Knights-Landing including a roofline study of code performance before and after a number of optimizations. We find that the GW method is particularly well-suited for many-core architectures due to the ability to exploit a large amount of parallelism over plane-wave components, band-pairs, and frequencies.

97 MATHEMATICS AND COMPUTING↗

HDF5 in the exascale era: Delivering efficient and scalable parallel I/O for exascale applications

Accurately modeling real-world systems requires scientific applications at exascale to generate massive amounts of data and manage data storage efficiently. However, parallel input and output (I/O) faces challenges due to new application workflows and the state-of-the-art memory, interconnect, and storage architectures considered in exascale designs. The storage hierarchy has expanded with node-local persistent memory, solid-state storage, and traditional disk and tape-based storage, thus requiring efficiency at each layer and much more efficient data movement among these layers. This paper discusses how the ExaHDF5 project improved the I/O performance and data management for exascale architectures by enhancing HDF5, a widely used parallel I/O library. The team developed an Asynchronous I/O Virtual Object Layer (VOL) connector that allowed overlapping I/O with computation. They also created a Cache VOL to complement asynchronous I/O by incorporating fast storage layers, such as burst buffer and node-local storage, into the parallel I/O workflow through caching and staging data. Additionally, the team enabled data aggregation and I/O at the node level by using a Subfiling Virtual File Driver (VFD). To demonstrate superior I/O performance with HDF5 at exascale, the ExaHDF5 team collaborated with several exascale applications. In this paper, we show I/O performance improvements for three applications: Cabana (a particle-based simulation library), EQSIM (a regional earthquake simulation software), and E3SM (a climate system modeling library).

Asynchronous I/Ol↗

Containers on Switches: A Cluster School Experience

Network switches, such as those from Arista and Mellanox, often have underutilized computational resources in the form of built-in processors and memory. By leveraging these untapped resources, we can optimize functionality and efficiency of computational cluster networks. Our research focuses on deploying containers directly onto these switches to execute various auxiliary tasks ranging from metric logging to system-wide management via post-boot configuration. By doing so, we can significantly enchance the capabilities of the cluster without the need for additional dedicated hardware. Our research involved five distinct scenarios where switch utilization could have a profound impact on HPC Clusters: run cloud-init services via link-local connection; configuring a Telegraf container to export metrics; deploying a caching proxy; creating a reconfigurable IPv6 DHCP/DNS provider for VLAN; and implementing a client detection with Magellan discovery. These scenarios were containerized with podman and docker, and tested both physically on the switch virtually on a QEMU VM both running SONiC OS. Testing and findings indicate that network switches can indeed be used for these scenarios. They offer a wide range of possibilities beyond these applications. They run as expected as containers on the switches, and although there were some minor issues, work-arounds were implemented. Overall, this is a positive result that can be further explored with more scenarios.

97 MATHEMATICS AND COMPUTING↗

Data Layout Transformation for Enhancing Data Locality on NUCA Chip Multiprocessors

With increasing numbers of cores, future CMPs (Chip Multi-Processors) are likely to have a tiled architecture with a portion of shared L2 cache on each tile and a bankinterleaved distribution of the address space. Although such an organization is effective for avoiding access hot-spots, it can cause a significant number of non-local L2 accesses for many commonly occurring regular data access patterns. In this paper we develop a compile-time framework for data locality optimization via data layout transformation. Using a polyhedral model, the program’s localizability is determined by analysis of its index set and array reference functions, followed by non-canonical data layout transformation to reduce non-local accesses for localizable computations. Simulation-based results on a 16-core 2D tiled CMP demonstrate the effectiveness of the approach. The developed program transformation technique is also useful in several other data layout transformation contexts.

99 GENERAL AND MISCELLANEOUS↗

Evaluating Operators in Deep Neural Networks for Improving Performance Portability of SYCL

SYCL is a portable programming model for heterogeneous computing, so it is important to obtain reasonable performance portability of SYCL. Towards the goal of better understanding and improving performance portability of SYCL for machine learning workloads, we have been developing benchmarks for basic operators in deep neural networks (DNNs). These operators could be offloaded to heterogeneous computing devices such as graphics processing units (GPUs) to speed up computation. In this work, we introduce the benchmarks, evaluate the performance of the operators on GPU-based systems, and describe the causes of the performance gap between the SYCL and Compute Unified Device Architecture (CUDA) kernels. We find that the causes are related to the utilization of the texture cache for read-only data, optimization of the memory accesses with strength reduction, shared local memory accesses, and register usage per thread. We hope that the efforts of developing benchmarks for studying performance portability will stimulate discussion and interactions within the community.

97 MATHEMATICS AND COMPUTING↗

Benchmarking Operators in Deep Neural Networks for Improving Performance Portability of SYCL

SYCL is a portable programming model for heterogeneous computing, so it is important to obtain reasonable performance portability of SYCL. Towards the goal of better understanding and improving performance portability of SYCL for machine learning workloads, we have been developing benchmarks for basic operators in deep neural networks (DNNs). These operators could be offloaded to heterogeneous computing devices such as graphics processing units (GPUs) to speed up computation. In this paper, we introduce the benchmarks, evaluate the performance of the operators on GPU-based systems, and describe the causes of the performance gap between the SYCL and Compute Unified Device Architecture (CUDA) kernels. We find that the causes are related to the utilization of the texture cache for read-only data, optimization of the memory accesses with strength reduction, use of local memory, and register usage per thread. We hope that the efforts of developing benchmarks for studying performance portability will stimulate discussion and interactions within the community.

Jin, Zheming [ORNL] (ORCID:000000027197780X)↗

Generating Performance Models for Irregular Applications

Many applications have irregular behavior --- non-uniform input data, input-dependent solvers, irregular memory accesses, unbiased branches --- that cannot be captured using today's automated performance modeling techniques. We describe new hierarchical critical path analyses for the \Palm model generation tool. To create a model's structure, we capture tasks along representative MPI critical paths. We create a histogram of critical tasks with parameterized task arguments and instance counts. To model each task, we identify hot instruction-level sub-paths and model each sub-path based on data flow, instruction scheduling, and data locality. We describe application models that generate accurate predictions for strong scaling when varying CPU speed, cache speed, memory speed, and architecture. We present results for the Sweep3D neutron transport benchmark; Page Rank on multiple graphs; Support Vector Machine with pruning; and PFLOTRAN's reactive flow/transport solver with domain-induced load imbalance.

analytical performance models↗

PCM-Based Durable Write Cache for Fast Disk I/O

Flash based solid-state devices (FSSDs) have been adopted within the memory hierarchy to improve the performance of hard disk drive (HDD) based storage system. However, with the fast development of storage-class memories, new storage technologies with better performance and higher write endurance than FSSDs are emerging, e.g., phase-change memory (PCM). Understanding how to leverage these state-of-the-art storage technologies for modern computing systems is important to solve challenging data intensive computing problems. In this paper, we propose to leverage PCM for a hybrid PCM-HDD storage architecture. We identify the limitations of traditional LRU caching algorithms for PCM-based caches, and develop a novel hash-based write caching scheme called HALO to improve random write performance of hard disks. To address the limited durability of PCM devices and solve the degraded spatial locality in traditional wear-leveling techniques, we further propose novel PCM management algorithms that provide effective wear-leveling while maximizing access parallelism. We have evaluated this PCM-based hybrid storage architecture using applications with a diverse set of I/O access patterns. Our experimental results demonstrate that the HALO caching scheme leads to an average reduction of 36.8% in execution time compared to the LRU caching scheme, and that the SFC wear leveling extends the lifetime of PCM by a factor of 21.6.

99 GENERAL AND MISCELLANEOUS↗

CSU Final Report on the Math/CS Institute CACHE: Communication-Avoiding and Communication-Hiding at the Extreme Scale

The CACHE project entails researching and developing new versions of numerical algorithms that result in data reuse that can be scheduled in a communication avoiding way. Since memory accesses take more time than any computation and require the most power, the focus on turning data reuse into data locality is critical to improving performance and reducing power usage in scientific simulations. This final report summarizes the accomplishments at Colorado State University as part of the CACHE project.

97 MATHEMATICS AND COMPUTING↗

Data Management in the Continuum: Cross-facility Object-based Data Transfers

Scientific workflows are evolving from relying on a monolithic storage subsystem at a single High-Performance Computing (HPC) facility to using geographically distributed file systems, repositories, and cloud storage. As a result, storing, accessing, transferring, and managing scientific data have become highly complex and prone to performance inefficiencies. This paper delves into these challenges by exploring an optimized end-to-end interface designed to seamlessly connect various local and remote storage systems, enabling efficient data movement of objects across HPC–Cloud and HPC–HPC environments. We showcase this capability through an object-focused data management runtime system, discuss the effects of relaxed consistency semantics in distributed object scenarios, and illustrate its application in an earthquake simulation workflow. Besides reducing the amount of data by selectively transferring regions of interest, our facility-local results achieved a speedup of 45 × over an optimized HDF5 usage and 15 × over the HDF5 with caching by using the new interface in PDC-XF.

Bez, Jean Luca↗

MULTI-CORE AND OPTICAL PROCESSOR RELATED APPLICATIONS RESEARCH AT OAK RIDGE NATIONAL LABORATORY

High-speed parallelization of common tasks holds great promise as a low-risk approach to achieving the significant increases in signal processing and computational performance required for next generation innovations in reconfigurable radio systems. Researchers at the Oak Ridge National Laboratory have been working on exploiting the parallelization offered by this emerging technology and applying it to a variety of problems. This paper will highlight recent experience with four different parallel processors applied to signal processing tasks that are directly relevant to signal processing required for SDR/CR waveforms. The first is the EnLight Optical Core Processor applied to matched filter (MF) correlation processing via fast Fourier transform (FFT) of broadband Dopplersensitive waveforms (DSW) using active sonar arrays for target tracking. The second is the IBM CELL Broadband Engine applied to 2-D discrete Fourier transform (DFT) kernel for image processing and frequency domain processing. And the third is the NVIDIA graphical processor applied to document feature clustering. EnLight Optical Core Processor. Optical processing is inherently capable of high-parallelism that can be translated to very high performance, low power dissipation computing. The EnLight 256 is a small form factor signal processing chip (5x5 cm2) with a digital optical core that is being developed by an Israeli startup company. As part of its evaluation of foreign technology, ORNL's Center for Engineering Science Advanced Research (CESAR) had access to a precursor EnLight 64 Alpha hardware for a preliminary assessment of capabilities in terms of large Fourier transforms for matched filter banks and on applications related to Doppler-sensitive waveforms. This processor is optimized for array operations, which it performs in fixed-point arithmetic at the rate of 16 TeraOPS at 8-bit precision. This is approximately 1000 times faster than the fastest DSP available today. The optical core performs the matrix-vector multiplications, where the nominal matrix size is 256x256. The system clock is 125MHz. At each clock cycle, 128K multiply-and-add operations per second (OPS) are carried out, which yields a peak performance of 16 TeraOPS. IBM Cell Broadband Engine. The Cell processor is the extraordinary resulting product of 5 years of sustained, intensive R&D collaboration (involving over $400M investment) between IBM, Sony, and Toshiba. Its architecture comprises one multithreaded 64-bit PowerPC processor element (PPE) with VMX capabilities and two levels of globally coherent cache, and 8 synergistic processor elements (SPEs). Each SPE consists of a processor (SPU) designed for streaming workloads, local memory, and a globally coherent direct memory access (DMA) engine. Computations are performed in 128-bit wide single instruction multiple data streams (SIMD). An integrated high-bandwidth element interconnect bus (EIB) connects the nine processors and their ports to external memory and to system I/O. The Applied Software Engineering Research (ASER) Group at the ORNL is applying the Cell to a variety of text and image analysis applications. Research on Cell-equipped PlayStation3 (PS3) consoles has led to the development of a correlation-based image recognition engine that enables a single PS3 to process images at more than 10X the speed of state-of-the-art single-core processors. NVIDIA Graphics Processing Units. The ASER group is also employing the latest NVIDIA graphical processing units (GPUs) to accelerate clustering of thousands of text documents using recently developed clustering algorithms such as document flocking and affinity propagation.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Multiple core computer processor with globally-accessible local memories

A multi-core computer processor including a plurality of processor cores interconnected in a Network-on-Chip (NoC) architecture, a plurality of caches, each of the plurality of caches being associated with one and only one of the plurality of processor cores, and a plurality of memories, each of the plurality of memories being associated with a different set of at least one of the plurality of processor cores and each of the plurality of memories being configured to be visible in a global memory address space such that the plurality of memories are visible to two or more of the plurality of processor cores.

97 MATHEMATICS AND COMPUTING↗