Parallel Data Processing Research Articles

Metaplastic and energy-efficient biocompatible graphene artificial synaptic transistors for enhanced accuracy neuromorphic computing

CMOS-based computing systems that employ the von Neumann architecture are relatively limited when it comes to parallel data storage and processing. In contrast, the human brain is a living computational signal processing unit that operates with extreme parallelism and energy efficiency. Although numerous neuromorphic electronic devices have emerged in the last decade, most of them are rigid or contain materials that are toxic to biological systems. In this work, we report on biocompatible bilayer graphene-based artificial synaptic transistors (BLAST) capable of mimicking synaptic behavior. The BLAST devices leverage a dry ion-selective membrane, enabling long-term potentiation, with ~50 aJ/µm2 switching energy efficiency, at least an order of magnitude lower than previous reports on two-dimensional material-based artificial synapses. The devices show unique metaplasticity, a useful feature for generalizable deep neural networks, and we demonstrate that metaplastic BLASTs outperform ideal linear synapses in classic image classification tasks. With switching energy well below the 1 fJ energy estimated per biological synapse, the proposed devices are powerful candidates for bio-interfaced online learning, bridging the gap between artificial and biological neural networks.

Efficient optical reservoir computing for parallel data processing

We propose and experimentally demonstrate an optical reservoir computing system in free space, using second-harmonic generation for nonlinear kernel functions and a scattering medium to enhance reservoir nodes interconnection. We test it for one-step and multi-step predication of Mackey-Glass time series with different input-mapping methods on a spatial light modulator. For one-step prediction, we achieve 1.8 × 10-3 normalized mean squared error (NMSE). For the multi-step prediction, we explore two different mapping methods: linear-combination and concatenation, achieving 16-step prediction with NMSE as low as 3.5 × 10-4. Robust and superior for multi-step prediction, our approach and design have potential for parallel data processing tasks such as video prediction, speech translation, and so on.

A robust high-performance electronic synapse based on epitaxial ferroelectric Hf0.5Zr0.5O2 films with uniform polarization and high Curie temperature

Ferroelectric tunnel junction (FTJ) is a promising emerging memristor for the artificial synapse in neuro-inspired computing, which has parallel data processing and low power consumption. The achievement of high-performance electronic synapses requires in-depth exploration of the correlation between the material properties and the device performances as well as the related physical mechanism, which are, however, still quite lacking. We demonstrate here a robust electronic synapse realized by epitaxial ferroelectric Hf0.5Zr0.5O2 (HZO) films with a high Curie temperature of 930 °C and a pristine highly uniform polarization. Based on the optimized ferroelectric HZO film and in-depth understanding of the FTJ mechanism, a robust and high-performance electronic synapse has been successfully realized with high ON/OFF ratio of >500, large continuous conductance regulation range of 1–250 nS and high reliability with the retention of >104 s. Such electronic synapses show good multilevel conductance modulations and synaptic behaviors, such as long-term potentiation (LTP), long-term depression (LTD) and spike-timing dependent plasticity (STDP). A simulated neural network with the synaptic characteristics indicates high recognition accuracy (93.7%) for MNIST database. These results pave a pathway to apply HZO based electronic synapses as the active block in future neuromorphic computing.

Analysis and Comparison of Different Approaches to Implementing a Network-Based Parallel Data Processing Algorithm

It is well known that network-based parallel data processing algorithms are well suited to implementation in reconfigurable hardware recurring to either Field-Programmable Gate Arrays (FPGA) or Programmable Systems-on-Chip (PSoC). The intrinsic parallelism of these devices makes it possible to execute several data-independent network operations in parallel. However, the approaches to designing the respective systems vary significantly with the experience and background of the engineer in charge. In this paper, we analyze and compare the pros and cons of using an embedded processor, high-level synthesis methods, and register-transfer low-level design in terms of design effort, performance, and power consumption for implementing a parallel algorithm to find the two smallest values in a dataset. This problem is easy to formulate, has a number of practical applications (for instance, in low-density parity check decoders), and is very well suited to parallel implementation based on comparator networks.

Data parallel processing in block streams for analysis of long-term series of ozone content in ground atmosphere

The article describes a new approach to the processing of long series of ozone monitoring data in the surface atmosphere called the Data Parallel Processing in Block Streams. The proposed method is based on splitting of a sequential series of initial data into blocks filled with surface ozone monitoring data for one day. Further, a chain is formed from such blocks, the length of which is determined by the total length of the monitoring process. Along this chain of blocks, parallel processing of the initial data is carried out, aimed at smoothing out fast fluctuations. The smoothed data is then used to determine the daily production of ozone due to photochemical reactions, its minimum nighttime levels, as well as the magnitude of nighttime maxima. The possibilities of the proposed approach are demonstrated using the analysis of surface ozone monitoring data in Moscow, Russia, in 2020. The prospects for further application of the developed method are discussed.

Research on Application of Computer Big Data Technology in Real-time Monitoring System of Bridge Structure Safety

The bridge structure safety real-time monitoring system based on big data evaluates the bridge safety status and real-time warning under the comprehensive consideration of various factors that affect the service performance of the bridge. The system platform uses a distributed file system with high fault tolerance and a parallel data processing engine with high computational efficiency. It has high reliability, availability and storage efficiency, and is easy to expand. Furthermore, the multi-factor analysis method is used to fully explore the implicit correlation between the real-time data of the various sensors of the bridge, and the bridge service performance evaluation model is accurately established by analysing the data correlation, and real-time warning of the bridge safety status is given. In addition, the system platform uses a variety of model verification methods to evaluate the effectiveness of the model to ensure the reliability of the big data analysis method.

An Approach to Data Processing for the Smart District Heating System

The article deals with the district heating systems transition to intelligent systems by developing a united information system and obtaining a high level of controllability of the entire system. During the implementation of automated control systems of district heating, a number of information tasks of the lower level are being introduced, including the data collection for thermal and hydraulic modes of operation for monitoring, operational management and analysis of the effectiveness. One of the problems of intelligent systems is data collection and its further storage and processing. Methods for data collection for real energy facilities are considered and the usage of multi-level system with the allocation of the upper level in the cloud storage has been proposed. In addition to the currently implemented data collection scheme in automated control systems, a generalized method of data acquisition with the introduction of duplicate streams has been proposed to ensure their integrity. The paper presents the approaches to identifying the collected data, ensuring the stability of the collection process, reliability of data storage and their integrity. Role-based security model with a dedicated single certification authority helps to protect data. Approaches to further processing of the collected data are shown, differing in the way of parallel data processing. The next stage of development is global monitoring systems that will be aimed to prompt response at all levels. The accumulated data will allow bringing the operating systems to a new level through the use of tools such as forecasting and simulation modeling, which will allow creating digital twins of heat supply systems. The proposed data collection system will perform forecasting and modeling at a higher level, and, as a result, help in the formation of more balanced management decisions.

Key Technologies of Media Big Data in-Depth Analysis System Based on 5G Platform

To meet the needs of large-scale users for personalized streaming media services with high speed, low delay, and high quality in a 5G mobile network environment, this paper studies the resource allocation mechanism of streaming media based on a 5G network from the perspective of user demand prediction, which can alleviate the pressure of mobile network, improve the utilization rate of streaming media resources and the quality of user service experience. The augmented reality visualization of large-scale social media data must rely on the computing power of distributed clusters. This paper constructs a distributed parallel processing framework in a high-performance cluster environment, which adopts a loosely coupled organizational structure. Each module can be combined, called, and expanded arbitrarily under the condition of following a unified interface. In this paper, the algebraic method of parallel computing algorithm is innovatively proposed to describe parallel processing tasks and organize and call large-scale data-parallel processing operators, which effectively supports the business requirements of large-scale parallel processing of large-scale spatial social media data and solves the bottleneck of large-scale spatial social media data-parallel processing.

Spark-Based Digital Factory Design

Abstract Big data processing often uses the paradigm of parallelism by computing directly on top of the distributed data storage. The existing big data workflows unify the data processing practices to utilize the cloud’s native computational potentials to offer advanced machine learning and BI capabilities. Spark is an open-source massively parallel in-memory data processing framework, the current state-of-the-art. The primary approach is to break down the job into granular-level executed tasks, enabling parallelization. In the discussed case study, through IoT – cloud solutions, the plant data can be converted into an analyzable form to let the farther machine learning modules produce added value. To maximize the efficiency of the processing and accumulation, cloud-based components are introduced. Based on the data insights, the appropriate operative actions can be taken. The cost and performance optimization methods were also discussed in the study. Through achieving higher degree of digitalization, the control over the production increased.

Collaborative Cluster Configuration for Distributed Data-Parallel Processing: A Research Overview

Many organizations routinely analyze large datasets using systems for distributed data-parallel processing and clusters of commodity resources. Yet, users need to configure adequate resources for their data processing jobs. This requires significant insights into expected job runtimes and scaling behavior, resource characteristics, input data distributions, and other factors. Unable to estimate performance accurately, users frequently overprovision resources for their jobs, leading to low resource utilization and high costs.In this paper, we present major building blocks towards a collaborative approach for optimization of data processing cluster configurations based on runtime data and performance models. We believe that runtime data can be shared and used for performance models across different execution contexts, significantly reducing the reliance on the recurrence of individual processing jobs or, else, dedicated job profiling. For this, we describe how the similarity of processing jobs and cluster infrastructures can be employed to combine suitable data points from local and global job executions into accurate performance models. Furthermore, we outline approaches to performance prediction via more context-aware and reusable models. Finally, we lay out how metrics from previous executions can be combined with runtime monitoring to effectively re-configure models and clusters dynamically.

Distributed machine learning for IoT

In the modern world, big data is used in machine learning, which is quite difficult to process on a single computer, so various methods for parallel processing of such data are being developed. But what about microcontrollers? In a cloud system, microcontrollers are often found, thanks to which they make pacification of various devices, and sometimes you have to work with big data. In microcontrollers, the memory is quite small and the processor is not as productive as on modern supercomputers. Therefore, many scientists propose various methods for parallel processing of big data for embedded systems, one of such methods is proposed by the author of this article.

Survey of Orion Disks with ALMA (SODA)

Context. Surveys of protoplanetary disks in nearby star-forming regions (SFRs) have provided important information on their demographics. However, due to their sample sizes, these surveys cannot be used to study how disk properties vary with the environment. Aims. We conduct a survey of the unresolved millimeter continuum emission of 873 protoplanetary disks identified by Spitzer in the L1641 and L1647 regions of the Orion A cloud. This is the largest such survey yet, allowing us to identify even weak trends in the median disk mass as a function of position in the cloud and cluster membership. The sample detection rates and median masses are also compared to those of nearby (<300 pc) SFRs. Methods. The sample was observed with the Atacama Large Millimeter/submillimeter Array (ALMA) at 225 GHz, with a median rms of 0.08 mJy beam−1, or 1.5 M⊕. The data were reduced and imaged using an innovative parallel data processing approach. Results. We detected 58% (502/873) of the observed disks. This includes 20 disks with dust masses >100 M⊕, and two objects associated with extended dust emission. By fitting a log-normal distribution to the data, we infer a median disk dust mass in the full sample of 2.2−0.2+0.2 M⊕. In L1641 and L1647, median dust masses are 2.1−0.2+0.2M⊕ and 2.6−0.5+0.4M⊕, respectively. Conclusions. The disk mass distribution of the full sample is similar to that of nearby low-mass SFRs at similar ages of 1–3 Myr. We find only weak trends in disk (dust) masses with galactic longitude and between the Young Stellar Object (YSO) clusters identified in the sample, with median masses varying by ≲50%. Differences in age may explain the median disk mass variations in our subsamples. Apart from this, disk masses are essentially constant at scales of ~100 pc. This also suggests that the majority of disks, even in different SFRs, are formed with similar initial masses and evolve at similar rates, assuming no external irradiation, with disk mass loss rates of ~10−8 M⊙ yr−1.

Scalable Artificial Intelligence for Earth Observation Data Using Hopsworks

This paper introduces the Hopsworks platform to the entire Earth Observation (EO) data community and the Copernicus programme. Hopsworks is a scalable data-intensive open-source Artificial Intelligence (AI) platform that was jointly developed by Logical Clocks and the KTH Royal Institute of Technology for building end-to-end Machine Learning (ML)/Deep Learning (DL) pipelines for EO data. It provides the full stack of services needed to manage the entire life cycle of data in ML. In particular, Hopsworks supports the development of horizontally scalable DL applications in notebooks and the operation of workflows to support those applications, including parallel data processing, model training, and model deployment at scale. To the best of our knowledge, this is the first work that demonstrates the services and features of the Hopsworks platform, which provide users with the means to build scalable end-to-end ML/DL pipelines for EO data, as well as support for the discovery and search for EO metadata. This paper serves as a demonstration and walkthrough of the stages of building a production-level model that includes data ingestion, data preparation, feature extraction, model training, model serving, and monitoring. To this end, we provide a practical example that demonstrates the aforementioned stages with real-world EO data and includes source code that implements the functionality of the platform. We also perform an experimental evaluation of two frameworks built on top of Hopsworks, namely Maggy and AutoAblation. We show that using Maggy for hyperparameter tuning results in roughly half the wall-clock time required to execute the same number of hyperparameter tuning trials using Spark while providing linear scalability as more workers are added. Furthermore, we demonstrate how AutoAblation facilitates the definition of ablation studies and enables the asynchronous parallel execution of ablation trials.

CMOS-compatible compute-in-memory accelerators based on integrated ferroelectric synaptic arrays for convolution neural networks.

Convolutional neural networks (CNNs) have gained much attention because they can provide superior complex image recognition through convolution operations. Convolution processes require repeated multiplication and accumulation operations, which are difficult tasks for conventional computing systems. Compute-in-memory (CIM) that uses parallel data processing is an ideal device structure for convolution operations. CIM based on two-terminal synaptic devices with a crossbar structure has been developed, but unwanted leakage current paths and the high-power consumption remain as the challenges. Here, we demonstrate integrated ferroelectric thin-film transistor (FeTFT) synaptic arrays that can provide efficient parallel programming and data processing for CNNs by the selective and accurate control of polarization in the ferroelectric layer. In addition, three-terminal FeTFTs can act as both nonvolatile memory and access device, which tackle issues from two-terminal devices. An integrated FeTFT synaptic array with parallel programming capabilities can perform convolution operations to extract image features with a high-recognition accuracy.

Design of Real-Time Digital Multi-Beamformer of Digital Array Antenna System for MFR

In this paper, we implement a digital multi-beamformer using FPGA(Field Programmable Gate Array) which has advantages in parallel and real-time data processing. This is accomplished through the use of not only high-speed data communication but also multiple beam forming, which is currently required by MFR(Multi Function Radar). As a result, the beamformer can process 24 Gbps throughput in real-time and form 5 digital beams at the same time. It is also compared to the results of Matlab simulations. We demonstrate how an implemented beamformer can be used in an MFR system by using a digital array antenna.

Challenge of the performance management of trust control systems with deep learning

Introduction. The significance of machine learning under the conditions of digital transformation of industry, and methods of implementing deep learning to provide the performance of trust management systems are considered. The necessity of using convolutional artificial neural networks for deep machine learning is determined. Various technologies and architectures for the implementation of artificial neural networks are briefly considered; a comparative analysis of their performance is carried out. The work objective is to study the need to develop new approaches to the architecture of computing machines for solving problems of deep machine learning in the trust management system implementation.Materials and Methods. In the context of digital transformation, the use of artificial intelligence reaches a new level. The technical implementation of artificial neural systems with deep machine learning is based on the use of one of three basic technologies: high performance computing (HPC) with parallel data processing, neuromorphic computing (NC), and quantum computing (QC).Results. Implementation models for deep machine learning, basic technologies and architecture of computing machines, as well as requirements for trust assurance in control systems using deep machine learning are analyzed. The problem of shortage of computation power for solving such problems is identified. None of the currently existing technologies can solve the full range of learning and impedance problems. The current level of technology does not provide information security and reliability of neural networks. The practical implementation of trust management systems with deep machine learning based on existing technologies for a significant part of the tasks does not provide a sufficient level of performance.Discussion and Conclusions. The study made it possible to identify the challenge of the computation power shortage for solving problems of deep machine learning. Through the analysis of the requirements for trust management systems, the external challenges of their implementation on the basis of existing technologies, and the need to develop new approaches to the computer architecture are determined.

Cloud-based virtualization environment for IoT-based WSN: solutions, approaches and challenges.

Internet of Things (IoT) is an ever-growing technology that enables advanced communication among millions of various devices to provide ubiquitous services without human intervention. The potential growth of electronic devices in sensing systems has led to the realization of IoT paradigm where applications depend on sensors to interact with the environment and collect data in a real-time scenario. Nowadays, smart applications require fast data acquisition, parallel processing, and dynamic resource sharing. Unfortunately, these requirements can not be supported efficiently with traditional Wireless Sensor Networks (WSN) due to the deficiency of computing resources and the lack of resource-sharing. Therefore, it is not recommended to develop innovative applications based on these constrained devices without further enhancement and improvement. Hence, this article explores a coeffective solution based on Cloud Computing and Virtualization Techniques to address these challenges. Cloud computing provides efficient computing resources and huge storage space, while the virtualization technique allows resources to be virtualized and shared between various applications. Integrating IoT-WSN with the Cloud-based Virtualization Environment will eliminate the drawbacks and limitations of conventional networks and facilitate the development of novel applications in a more flexible way. Furthermore, this article reviews the recent trends in IoT-WSN, virtualization techniques, and cloud computing. Also, we present the integration process of sensor networks with Cloud-based Virtualization and propose a new general architecture view for the Sensor-Cloud paradigm, and discuss its key elements, basic principles, lifecycle operation, and outline its advantages and disadvantages. Finally, we review the state-of-the-art, present the major challenges, and suggest future work directions.

OVERVIEW OF OPTIMIZATION METHODS FOR PRODUCTIVITY OF THE ETL PROCESS

One of the important aspects in management and acceleration of processes, operations in databases and data warehouses is ETL processes, the process of extracting, transforming and loading data. These processes without optimizing, a realization data warehouse project is costly, complex, and time-consuming. This paper provides an overview and research of methods for optimizing the performance of ETL processes; that the most important indicator of ETL system's operation is the time and speed of data processing is shown. The issues of the generalized structure of ETL process flows are considered, the architecture of ETL process optimization is proposed, and the main methods of parallel data processing in ETL systems are presented, those methods can improve its performance. The most relevant today of the problem is performance of ETL processes for data warehouses is considered in detail.

A joint computational and resource allocation model for fast parallel data processing in fog computing

A joint computational and resource allocation model for fast parallel data processing in fog computing

A two-level formal model for Big Data processing programs

• A high-level model for iterative and non-iterative large-scale data processing programs. • Formalizing data flow with Petri Nets and data processing operations with Monoid Algebra. • Application of the model to formalize mutation operators for mutation testing in data flow programs. • Proposal of new mutation operators for iterative data flow programs. This paper proposes a model for specifying data flow-based parallel data processing programs agnostic of target Big Data processing frameworks. The paper focuses on the formal abstract specification of non-iterative and iterative programs, generalizing the strategies adopted by data flow Big Data processing frameworks. The proposed model relies on Monoid Algebra and Petri Nets to abstract Big Data processing programs in two levels: a higher level representing the program data flow and a lower level representing data transformation operations (e.g., filtering, aggregation, join). We extend the model for data processing programs proposed in [1] , for modeling iterative data processing programs. The general specification of these programs implemented by data flow-based parallel programming models is essential given the democratization of iterative and greedy Big Data analytics algorithms. Indeed, these algorithms call for revisiting parallel programming models to express iterations. The paper gives a comparative analysis of the iteration strategies proposed by Apache Spark, DryadLINQ, Apache Beam, and Apache Flink. It discusses how the model achieves to generalize these strategies.