Multi-core Research Articles

Revisiting Database Indexing for Parallel and Accelerated Computing: A Comprehensive Study and Novel Approaches

While the importance of indexing strategies for optimizing query performance in database systems is widely acknowledged, the impact of rapidly evolving hardware architectures on indexing techniques has been an underexplored area. As modern computing systems increasingly leverage parallel processing capabilities, multi-core CPUs, and specialized hardware accelerators, traditional indexing approaches may not fully capitalize on these advancements. This comprehensive experimental study investigates the effects of hardware-conscious indexing strategies tailored for contemporary and emerging hardware platforms. Through rigorous experimentation on a real-world database environment using the industry-standard TPC-H benchmark, this research evaluates the performance implications of indexing techniques specifically designed to exploit parallelism, vectorization, and hardware-accelerated operations. By examining approaches such as cache-conscious B-Tree variants, SIMD-optimized hash indexes, and GPU-accelerated spatial indexing, the study provides valuable insights into the potential performance gains and trade-offs associated with these hardware-aware indexing methods. The findings reveal that hardware-conscious indexing strategies can significantly outperform their traditional counterparts, particularly in data-intensive workloads and large-scale database deployments. Our experiments show improvements ranging from 32.4% to 48.6% in query execution time, depending on the specific technique and hardware configuration. However, the study also highlights the complexity of implementing and tuning these techniques, as they often require intricate code optimizations and a deep understanding of the underlying hardware architecture. Additionally, this research explores the potential of machine learning-based indexing approaches, including reinforcement learning for index selection and neural network-based index advisors. While these techniques show promise, with performance improvements of up to 48.6% in certain scenarios, their effectiveness varies across different query types and data distributions. By offering a comprehensive analysis and practical recommendations, this research contributes to the ongoing pursuit of database performance optimization in the era of heterogeneous computing. The findings inform database administrators, developers, and system architects on effective indexing practices tailored for modern hardware, while also paving the way for future research into adaptive indexing techniques that can dynamically leverage hardware capabilities based on workload characteristics and resource availability.

Efficient parallel solver for rarefied gas flow using GSIS

Recently, the general synthetic iterative scheme (GSIS) has been proposed to find the steady-state solution of the Boltzmann equation in the whole range of gas rarefaction, where its fast-converging and asymptotic-preserving properties lead to the significant reduction of iteration numbers and spatial cells in the near-continuum flow regime. However, the efficiency and accuracy of GSIS have only been demonstrated in two-dimensional problems with small numbers of spatial cells and discrete velocities. Here, a large-scale parallel computing strategy is designed to extend the GSIS to three-dimensional flow problems, including the supersonic flows which are usually difficult to solve by the discrete velocity method. Since the GSIS involves the calculation of the mesoscopic kinetic equation which is defined in six-dimensional phase-space, and the macroscopic high-temperature Navier–Stokes–Fourier equations in three-dimensional physical space, the proper partition of the spatial and velocity spaces, and the allocation of CPU cores to the mesoscopic and macroscopic solvers, are the keys to improving the overall computational efficiency. These factors are systematically tested to achieve optimal performance, up to 100 billion spatial and velocity grids. For hypersonic flows around the Apollo reentry capsule, the X38-like vehicle, and the space station, our parallel solver can obtain the converged solution within one hour.

Responses of Ecosystem Services to Land Use/Cover Changes in Rapidly Urbanizing Areas: A Case Study of the Shandong Peninsula Urban Agglomeration

The rapid expansion of built-up land, a hallmark of accelerated urbanization, has emerged as a pivotal factor contributing to regional climate change and the degradation of ecosystem functions. The decline in ecosystem service value (ESV) has consequently garnered significant attention in global sustainable development research. The Shandong Peninsula urban agglomeration is crucial for promoting the construction of the Yellow River Economic Belt in China, with its ecological status increasingly gaining prominence. This study investigated the ESV response to land use/cover change (LUCC) through the elasticity coefficient in order to analyze the degree of disturbance caused by land use activities on ecosystem functions in the Shandong Peninsula urban agglomeration. This analysis was based on the examination of LUCC characteristics and ESV from 1990 to 2020. The findings reveal that (1) the Shandong Peninsula urban agglomeration experienced a continuous increase in the proportion of built-up land from 1990 to 2020, alongside a highly complex transfer between different land use types, characterized by diverse transfer trajectories. The most prominent features were noted to be the rapid expansion of built-up land and the simultaneous decline in agricultural land. (2) The analysis of four landscape pattern indices, encompassing Shannon’s diversity index, indicates that the continuous development of urbanization has led to increased fragmentation in land use and decreased connectivity. However, obvious spatial distribution differences exist among different districts and counties. (3) The ESV was revised using the normalized difference vegetation index, revealing a slight decrease in the total ESV of the Shandong Peninsula urban agglomeration. However, significant differences were observed among districts and counties. The number of counties and districts exhibiting low and high ESVs continuously increased, whereas those with intermediate levels generally remained unchanged. (4) The analysis of the elasticity coefficient reveals that LUCC exerts a substantial disturbance and influence on ecosystem services, with the strongest disturbance ability occurring from 2000 to 2010. The elasticity coefficient exhibits obvious spatial heterogeneity across both the entire urban agglomeration and within individual cities. Notably, Qingdao and Jinan, the dual cores of the Shandong Peninsula urban agglomeration, exhibit markedly distinct characteristics. These disparities are closely related to their development foundations in 1990 and their evolution over the past 30 years. The ESV response to LUCC displays significant variation across different time periods and spatial locations. Consequently, it is imperative to formulate dynamic management policies on the basis of regional characteristics. Such policies aim to balance social and economic development while ensuring ecological protection, thereby promoting the social and economic advancement and ecological environment preservation of the Shandong Peninsula urban agglomeration.

Inference latency prediction for CNNs on heterogeneous mobile devices and ML frameworks

Due to the proliferation of inference tasks on mobile devices, state-of-the-art neural architectures are typically designed using Neural Architecture Search (NAS) to achieve good tradeoffs between machine learning accuracy and inference latency. While measuring inference latency of a huge set of candidate architectures during NAS is not feasible, latency prediction for mobile devices is challenging, because of hardware heterogeneity, optimizations applied by machine learning frameworks, and diversity of neural architectures. Motivated by these challenges, we first quantitatively assess the characteristics of neural architectures (specifically, convolutional neural networks for image classification), ML frameworks, and mobile devices that have significant effects on inference latency. Based on this assessment, we propose an operation-wise framework which addresses these challenges by developing operation-wise latency predictors and achieves high accuracy in end-to-end latency predictions, as shown by our comprehensive evaluations on multiple mobile devices using multicore CPUs and GPUs. To illustrate that our approach does not require expensive data collection, we also show that accurate predictions can be achieved on real-world neural architectures using only small amounts of profiling data.

Refractive insensitive directional bend sensor based on specialty microstructure optical fiber with dumbbell shape core

With the development of information society, more and more special occasions are calling for a wide range of sensors with special properties. In this work, a refractive index insensitive directional bend sensor is proposed based on a specialty microstructure optical fiber with dumbbell shape core, which works as a dual core fiber (DCF). Firstly, this specialty microstructure optical fiber with dumbbell shape core has been fabricated with self-pressurised drawing technique. Then, it has been constructed as a Mach-Zehnder interferometer (MZI) sensor. Consequently, its bend sensing characteristics has been investigated from both theory and experiment. The response of the proposed sensor to the bending evidently displays the directional dependence. In the bending along x direction, the interference peak wavelength is almost insensitive to the curvature. But the transmission is sensitive, which almost obeys the exponential function. Correspondingly, in the bending along y direction the wavelength almost linearly varies with curvature, giving out an averaged sensitivity of 2.44 nm/m−1 in the curvature range of 0–3.79 m−1. But the peak transmission fluctuates without typical principle. In addition, the response of the proposed sensor to the refractive index variation has been checked. It is found that the peak wavelength is almost insensitive to the change of refractive index from 1.33 to 1.45, and the peak transmission fluctuates without clear principle. All these results demonstrate that the proposed sensor has great potential for future internet of things (IoT) applications.

Direct 3D-printed ring-resonator photonic circuit on a dual core fiber tip for remote sensing applications.

On-chip optical sensors using ring- and disk-resonators have many potential sensing applications, yet robust and efficient fiber-to-chip coupling and the differing form factor between the two pose deployment challenges. To resolve this, we 3D-printed a ring-resonator onto the tip of a dual-core fiber and demonstrate its use as a remote temperature sensor. The fiber-tip optical circuit is fabricated using direct laser writing (DLW) with two-photon absorption photopolymer material IP-Dip, forming micrometer-scale waveguide cores having a refractive index of 1.53 with a surrounding air cladding. We connect the two-fiber cores by a printed bus-waveguide, utilizing total internal reflection mirrors, allowing light launched into one core to be guided back to the other core. Furthermore, a DLW printed racetrack resonator evanescently coupled to the bus waveguide (Q ∼ 3000) imposes spectral dips on resonance wavelengths. Light sent down into one core is interrogated upon return from the second core, all from the distal end of the sensor. When the sensing end's temperature is varied, we find a sensitivity of 78 pm/K, due to the polymer's thermo-optic index variation. The ring-resonator could be functionalized for other sensing applications.

PowerTrain: Fast, generalizable time and power prediction models to optimize DNN training on accelerated edges

Accelerated edge devices, like Nvidia’s Jetson with 1000+ CUDA cores, are increasingly used for DNN training and federated learning, rather than just for inferencing workloads. A unique feature of these compact devices is their fine-grained control over CPU, GPU, memory frequencies, and active CPU cores, which can limit their power envelope in a constrained setting while throttling the compute performance. Given this vast 10k+ parameter space, selecting a power mode for dynamically arriving training workloads to exploit power–performance trade-offs requires costly profiling for each new workload, or is done ad hoc. We propose PowerTrain, a transfer-learning approach to accurately predict the power and time that will be consumed when we train a given DNN workload (model + dataset) using any specified power mode (CPU/GPU/memory frequencies, core-count). It requires a one-time offline profiling of 1000s of power modes for a reference DNN workload on a single Jetson device (Orin AGX) to build Neural Network (NN) based prediction models for time and power. These NN models are subsequently transferred (retrained) for a new DNN workload, or even a different Jetson device, with minimal additional profiling of just 50 power modes to make accurate time and power predictions. These are then used to rapidly construct the Pareto front and select the optimal power mode for the new workload, e.g., to minimize training time while meeting a power limit. PowerTrain’s predictions are robust to new workloads, exhibiting a low MAPE of <6% for power and <15% for time on six new training workloads (MobileNet, YOLO, BERT, LSTM, etc.) for up to 4400 power modes, when transferred from a ResNet reference workload on Orin AGX. It is also resilient when transferred to two entirely new Jetson devices (Xavier AGX and Jetson Orin Nano) with prediction errors of <14.5% and <11%. These outperform baseline predictions by more than 10% and baseline optimizations by up to 45% on time and 88% on power.

Benchmarking parallel programming for single-board computers

Within the computing continuum, SBCs (single-board computers) are essential in the Edge and Fog, with many featuring multiple processing cores and GPU accelerators. In this way, parallel computing plays a crucial role in enabling the full computational potential of SBCs. However, selecting the best-suited solution in this context is inherently complex due to the intricate interplay between PPI (parallel programming interface) strategies, SBC architectural characteristics, and application characteristics and constraints. To our knowledge, no solution presents a combined discussion of these three aspects. To tackle this problem, this article aims to provide a benchmark of the best-suited parallelism PPIs given a set of hardware and application characteristics and requirements. Compared to existing benchmarks, we introduce new metrics, additional applications, various parallelism interfaces, and extra hardware devices. Therefore, our contributions are the methodology to benchmark parallelism on SBCs and the characterization of the best-performing parallelism PPIs and strategies for given situations. We are confident that parallel computing will be mainstream to process edge and fog computing; thus, our solution provides the first insights regarding what kind of application and parallel programming interface is the most suited for a particular SBC hardware.

Self-centering dual rocking core system with viscous dampers in additional rocking sections

To mitigate the high mode effects and reduce the force requirements of the structural members in the multi-story self-centering dual rocking core (SDRC) system, this paper proposes the multiple self-centering dual rocking core system with viscous dampers in additional rocking sections (MSDRCV system). The sketch and ideal working mechanisms of the MSDRCV system are interpreted. The additional rocking sections are set to decrease the force requirements of structural members and viscous dampers are included to mitigate high mode responses of the additional rocking sections, introducing seismic performance enhancement. Comparative analyses on a 9-story MSDRCV system with viscous dampers in rocking sections, a 9-story multiple self-centering dual rocking core system without viscous dampers in rocking sections (MSDRC system) and a 9-story SDRC system were carried out. The MSDRCV and MSDRC system can realize the reduction of the force demands in structural members in contrast to the SDRC system. Nevertheless, the high mode responses caused the uncontrollable displacement concentration in the MSDRC system and hence aggravated the seismic performance under rare seismic events. The analysis results in the frequency domain confirmed that the viscous dampers in the MSDRCV system can efficiently control the high-mode responses. The effects of the design parameters of the viscous dampers on the peak responses of the MSDRCV systems were comprehensively studied via parametric analyses. The results offer guidance for the viscous dampers design.

Accelerated dimensionality reduction of single-cell RNA sequencing data with fastglmpca.

Motivated by theoretical and practical issues that arise when applying Principal Components Analysis (PCA) to count data, Townes et al introduced "Poisson GLM-PCA", a variation of PCA adapted to count data, as a tool for dimensionality reduction of single-cell RNA sequencing (RNA-seq) data. However, fitting GLM-PCA is computationally challenging. Here we study this problem, and show that a simple algorithm, which we call "Alternating Poisson Regression" (APR), produces better quality fits, and in less time, than existing algorithms. APR is also memory-efficient, and lends itself to parallel implementation on multi-core processors, both of which are helpful for handling large single-cell RNA-seq data sets. We illustrate the benefits of this approach in two published single-cell RNA-seq data sets. The new algorithms are implemented in an R package, fastglmpca. The fastglmpca R package is released on CRAN for Windows, macOS and Linux, and the source code is available at github.com/stephenslab/fastglmpca under the open source GPL-3 license. Scripts to reproduce the results in this paper are also available in the GitHub repository. mstephens@uchicago.edu. Supplementary data are available on BioRxiv online.

Analysis of High-Performance Parallel Computing using TOPSIS Method

High-performance parallel computing involves the simultaneous execution of multiple tasks or processes, orchestrated to achieve improved computational speed and efficiency. This approach leverages the power of parallelism, exploiting both multi-core CPUs and GPUs, distributed computing clusters, and specialized hardware accelerators. The fundamental idea is to divide a task into smaller sub-tasks that can be executed concurrently, thereby reducing processing time and enhancing overall performance. High-performance parallel computing is a transformative approach that enables us to tackle computationally intensive tasks efficiently. This abstract highlight its significance in contemporary computing and sets the stage for further exploration of the intricacies and innovations within this dynamic field. Researchers and practitioners continue to push the boundaries of what is achievable, making high-performance parallel computing a cornerstone of modern computational science and technology. High-performance parallel computing research is of paramount significance due to its transformative impact across diverse fields. It empowers scientists to tackle complex problems that were once computationally intractable, unlocking new frontiers in scientific discovery. It drives innovation in engineering and design, optimizing product development and manufacturing processes across industries. In healthcare, it accelerates genomics research and drug discovery, offering hope for improved medical treatments. Financial institutions rely on it for data analysis and risk assessment, shaping the global economy. Weather forecasting and environmental modelling are enhanced, aiding disaster preparedness and conservation efforts. In the digital age, parallel computing underpins artificial intelligence, enabling advancements in natural language processing and machine learning. Furthermore, it has vital applications in national security, space exploration, and materials science. In essence, high-performance parallel computing research serves as the backbone of technological progress, fostering innovation, efficiency, and problem-solving across a wide spectrum of disciplines, ultimately shaping the future of our world. TOPSIS, this method involves evaluating the geometric distance between each alternative solution and two reference solutions: the positive ideal solution and the negative ideal solution. The underlying principle of TOPSIS assumes that the criteria being assessed are of an ascending nature, where larger values represent better performance. To account for disparate dimensions or scales among the criteria, normalization is often employed within the TOPSIS framework. From the result Scatter- free imaging is got the first rank and object-scatter imaging is having the lowest rank.

Software architecture and implementation based on multi-core digital signal processors

Abstract Embedded detection systems are a critical research topic in intelligent monitoring equipment. In response to the current challenges in balancing real-time performance, accuracy, and stability in embedded hardware detection systems, this paper proposes a software architecture system based on multi-core digital signal processing (DSP), incorporating the image detection Canny algorithm into this system. Initially, the paper discusses the selection criteria for multi-core DSPs and then introduces a multi-core self-check function that can effectively ensure system stability. It also proposes a multi-core communication handshake protocol that significantly enhances the system’s real-time performance. Furthermore, an improved multi-core master-slave architecture is presented, which includes the addition of a timer to the conventional architecture, providing a temporal guarantee for system operation. Finally, the Canny algorithm is selected and applied to this software architecture, with parallel acceleration processing applied to parts of the algorithm. Operational tests have demonstrated that the system maintains stable internal data flow, with its detection accuracy and speed meeting the set requirements.

MP-ORAM: A Novel ORAM Design for Multicore Processor Systems

MP-ORAM: A Novel ORAM Design for Multicore Processor Systems

New generalizations of the core and dual core inverses

New generalizations of the core and dual core inverses

Optimizing High-Speed Serial Links for Multicore Processors and Network Interfaces

In modern computing environments, high-speed serial links have become a critical component for ensuring efficient data transfer between multicore processors and network interfaces. These links, characterized by their ability to transmit data at very high rates over single or multiple lanes, are essential for meeting the increasing bandwidth demands of contemporary applications. The optimization of these serial links is crucial for maintaining performance, reliability, and energy efficiency in systems that leverage multicore processors and advanced network interfaces.This paper explores the key strategies for optimizing high-speed serial links in the context of multicore processors and network interfaces. We begin by examining the architectural considerations and design principles that influence the performance of serial links, including signal integrity, power consumption, and thermal management. We also discuss the impact of link speed and data encoding techniques on overall system efficiency.

Research and optimization of task scheduling algorithm based on heterogeneous multi-core processor

Research and optimization of task scheduling algorithm based on heterogeneous multi-core processor

Efficient Velvet-Noise Convolution in Multicore Processors

Velvet noise, a sparse pseudo-random signal, finds valuable applications in audio engineering, such as artificial reverberation, decorrelation filtering, and sound synthesis. These applications rely on convolution operations whose computational requirements depend on the length, sparsity, and bit resolution of the velvet-noise sequence used as filter coefficients. Given the inherent sparsity of velvet noise and its occasional restriction to a few distinct values, significant computational savings can be achieved by designing convolution algorithms that exploit these unique properties. This paper shows that an algorithm called the transposed double-vector filter is the most efficient way of convolving velvet noise with an audio signal. This method optimizes access patterns to take advantage of the processor’s fast caches. The sequential sparse algorithm is shown to be always faster than the dense one, and the speedup is linearly dependent on sparsity. The paper also explores the potential for further speedup on multicore platforms through parallelism and evaluate the impact of data encoding, including 16-bit and 32-bit integers and 32-bit floating-point representations. The results show that using the fastest implementation of a long velvet-noise filter, it is possible to process more than 40 channels of audio in real time using the quad-core processor of a modern system-on-chip.

Numerical investigation of lattice Boltzmann method-based large-eddy simulation in non-isothermal enclosed cavity airflow

This study implemented the lattice Boltzmann method-based large-eddy simulation (LBM-LES) in a wall-heated cavity turbulent flow problem with the Rayleigh number reaching 5.33 × 109. The effects of crucial parameters in LBM temperature simulation, including discrete time intervals, discrete external force models by temperature effect, and temperature discrete velocity schemes, were investigated to obtain the suggested values. In addition, the simulations' accuracy and computational efficiency were compared with those of FVM-LES. LBM-LES generally agreed with experiments, and the accuracy was comparable to FVM-LES. However, exceedingly small discrete time intervals induced numerical oscillations in velocity and temperature at high frequencies, yielding significant errors. The accuracy of the discrete external force model satisfied Guo model > primitive model > He model. For temperature particles' discrete velocity scheme, D3Q7 was already sufficiently accurate, and higher schemes (D3Q19 and D3Q27) did not increase the accuracy further. LBM-LES and FVM-LES were concordant with respect to predicting various physical quantities, and the accuracy of LBM-LES in the near-wall region was slightly higher than that of the FVM-LES. Single-core and multicore computational speeds of LBM-LES were 4–8 fold higher than FVM-LES. The multicore parallelism of LBM-LES also exceeded FVM-LES and became more evident as CPU cores increased.

Lavender: An Efficient Resource Partitioning Framework for Large-Scale Job Colocation

Workload consolidation is a widely used approach to enhance resource utilization in modern data centers. However, the concurrent execution of multiple jobs on a shared server introduces contention for essential shared resources such as CPU cores, Last Level Cache, and memory bandwidth. This contention negatively impacts job performance, leading to significant degradation in throughput. To mitigate resource contention, effective resource isolation techniques at the software or hardware level can be employed to partition the shared resources among colocated jobs. However, existing solutions for resource partitioning often assume a limited number of jobs that can be colocated, making them unsuitable for scenarios with a large-scale job colocation due to several critical challenges. In this study, we propose Lavender, a framework specifically designed for addressing large-scale resource partitioning problems. Lavender incorporates several key techniques to tackle the challenges associated with large-scale resource partitioning, ensuring efficiency, adaptivity, and optimality. We conducted comprehensive evaluations of Lavender to validate its performance and analyze the reasons for its advantages. The experimental results demonstrate that Lavender significantly outperforms state-of-the-art baselines. Lavender is publicly available at https://github.com/yanxiaoqi932/OpenSourceLavender.

The Digital Media Art and Design System under the Integration of Computer Big Data ChatGPT Technology

Traditional digital media art and design systems take a long time to run and cannot handle a large number of objects simultaneously. Therefore, the author proposes a digital media art and design method based on ChatGPT technology. ChatGPT technology is an important part of networking.Prepare to provide conference support for various applications such as customer service, personal assistant, robotic interview, and so on. ChatGPT can not only answer the question that is set, but also solve the question openly.The comprehensiveness and intelligence of ChatGPT have attracted widespread attention since its launch. The hardware system adopts dual core TMS320C6657 and DDR3Thel1333 series processors as the core circuit, together with high speed external storage, to expand the image transmission line of multi-data; System software uses module architecture to design image modules, video modules, and voice modules, all of which are integrated into the core of the system as modules. In this experiment, 10 images were transferred and the running time was tested. Experimental results showed that, the traditional digital media art and design system had three errors, which took more time to run.