High-performance Resources Research Articles

Efficient multi-task progressive learning for semantic segmentation and disparity estimation

Scene understanding is an important area in robotics and autonomous driving. To accomplish these tasks, the 3D structures in the scene have to be inferred to know what the objects and their locations are. To this end, semantic segmentation and disparity estimation networks are typically used, but running them individually is inefficient since they require high-performance resources. A possible solution is to learn both tasks together using a multi-task approach. Some current methods address this problem by learning semantic segmentation and monocular depth together. However, monocular depth estimation from single images is an ill-posed problem. A better solution is to estimate the disparity between two stereo images and take advantage of this additional information to improve the segmentation. This work proposes an efficient multi-task method that jointly learns disparity and semantic segmentation. Employing a Siamese backbone architecture for multi-scale feature extraction, the method integrates specialized branches for disparity estimation and coarse and refined segmentations, leveraging progressive task-specific feature sharing and attention mechanisms to enhance accuracy for solving both tasks concurrently. The proposal achieves state-of-the-art results for joint segmentation and disparity estimation on three distinct datasets: Cityscapes, TrimBot2020 Garden, and S-ROSeS, using only 1/3 of the parameters of previous approaches.

A multi-objective grey-wolf optimization based approach for scheduling on cloud platforms

A cloud computing environment processes user workloads or tasks by exploiting its high performance computational, storage, of reducing and network resources. The virtual machines in the cloud environment are allocated to tasks with the aim of reducing overall execution time. The use of high performance resources incurs monetary costs, as well as high power consumption. The heuristic based approaches implemented for scheduling tasks are unable to cope with the complexity of optimizing multiple parameters. In this paper, we propose a multi-objective grey-wolf optimization based algorithm for scheduling tasks on cloud platforms. The proposed algorithm targets to minimize schedule length (overall execution time), energy consumption, and monetary cost required for executing tasks. For optimization, the algorithm incorporates steps that are performed iteratively for mimicking the behavior of grey wolves attacking their prey. It uses discrete values for positioning wolves for encircling and attacking the prey. The assignment of tasks to virtual machines is performed using the solution found after multi-objective optimization that incorporates weighted sorting for arranging solutions. Our experimentation performed using the CloudSim framework shows that the proposed algorithm outperforms other algorithms with performance improvement ranging from 3.98% to 16.07%, while considering the schedule length, monetary cost, and energy consumption.

Data Center Efficiency Model: A New Approach and the Role of Artificial Intelligence

Bioinformatics technologies play a significant and growing role in life science research, and as these technologies develop, so does the complexity of data. The challenge of biological data growth has given rise to a number of bioinformatics data centers that offer services and solutions ranging from large-scale biosystems analyze that accounts for entire OMICs to nanoscale experiments where molecular modeling can provide insight o structure and dynamics of molecular complexes of biological components. Obviously, this kind of research requires a highly specialized level of computational and statistical expertise, as well as high-performance resources. The importance of information technology is growing, as is the use of computer information systems throughout the world. There are more and more specialized data centers and they consume more energy. The development of new strategies for energy efficiency of data centers is becoming relevant. These strategies aim to reduce the amount of energy consumed by data centers and their environmental impact without sacrificing performance. The article examines performance metrics, proposes a new method for data center energy efficiency, and discusses the role of artificial intelligence techniques in achieving these goals.

Binder-free Fe nano oxides decoration for stimulating power generation in microbial fuel cell: Effects on electrode substrates and function mechanism

Nanomaterials are ideal for electrode modification to improve the power output of microbial fuel cells (MFCs). The nanomaterials decoration usually needs a conductive binder which faces the problem of high cost and low conductivity, significantly constraining further application. This study successfully fabricated the binder-free Fe nano oxides modified anodes through the solvothermal method. MFC assembled with carbon cloth/Fe nano oxides (M−CC/Fe) achieved the highest maximum power density (471.6 mW m−2), dramatically higher than that with carbon cloth (M−CC, 150.4 mW m−2), Ni foam/Fe nano oxides (M−NF/Fe, 130.6 mW m−2) and Ni foam (M−NF, 16.5 mW m−2). The electrochemical catalyzing activity, extracellular electron transfer rate, COD removal efficiency and coulombic efficiency were all improved after Fe nano oxides decoration in MFCs. Fe nano oxides enriched exoelectrogens/iron-reducing bacteria like Geobacter and Comamonas and showed a mutual benefit relationship with them. The formation of stable Fe nano oxide cubes significantly enhanced the electroactive biofilm development and alleviated the corrosion of the electrode substrate. The complementary merits of the carbon-based electrode and Fe nano oxides performed better in stimulating power generation in MFCs. These results provided knowledge of in-situ nanomaterials decoration for advanced anodes, providing suggestions for high-performance resource recovery from wastewater.

Modelo de liderazgo saludable integral y su desarrollo como competencia psicosocial adaptativa en ambientes organizacionales complejos

The prevention of negative leadership as a psychosocial risk factor at work has shown relevant results in organizations from the theoretical framework of Occupational Health Psychology. The aim of this paper is to propose a Healthy-Extended Leadership Model with three elements: 1) a comprehensive definition; 2) a model with five dimensions: a) Constructive Social Relationships; b) Support for Training and Adaptive Development; c) Extended Health and Wellbeing Care; d) Provision of Resources for High Performance; e) Promotion of Favorable and Productive Workplaces; and 3) it is developed as a competency for its application in organizations. As conclusions, the propose model: can contribute to the balance between health and outstanding performance; contribute to the creation of favorable work environments; and, together with other protective psychosocial factors and competencies, have a larger impact on the development of healthy workers, organizations, and societies.

FSpot: Fast and Efficient Video Encoding Workloads Over Amazon Spot Instances

HTTP Adaptive Streaming (HAS) of video content is becoming an undivided part of the Internet and accounts for most of today's network traffic. Video compression technology plays a vital role in efficiently utilizing network channels, but encoding videos into multiple representations with selected encoding parameters is a significant challenge. However, video encoding is a computationally intensive and time-consuming operation that requires high-performance resources provided by on-premise infrastructures or public clouds. In turn, the public clouds, such as Amazon elastic compute cloud (EC2), provide hundreds of computing instances optimized for different purposes and clients’ budgets. Thus, there is a need for algorithms and methods for optimized computing instance selection for specific tasks such as video encoding and transcoding operations. Additionally, the encoding speed directly depends on the selected encoding parameters and the complexity characteristics of video content. In this paper, we first benchmarked the video encoding performance of Amazon EC2 spot instances using multiple ×264 codec encoding parameters and video sequences of varying complexity. Then, we proposed a novel fast approach to optimize Amazon EC2 spot instances and minimize video encoding costs. Furthermore, we evaluated how the optimized selection of EC2 spot instances can affect the encoding cost. The results show that our approach, on average, can reduce the encoding costs by at least 15.8% and up to 47.8% when compared to a random selection of EC2 spot instances.

Scoring and Dynamic Hierarchy-Based NSGA-II for Multiobjective Workflow Scheduling in the Cloud

Cloud computing becomes a promising technology to reduce computation cost by providing users with elastic resources and application-deploying environments as a pay-per-use model. More scientific workflow applications have been moved or are being migrated to the cloud. Scheduling workflows turns to the main bottleneck for increasing resource utilization and quality of service (QoS) for users. This work formulates workflow scheduling as multiobjective optimization problems and proposes a Scoring and Dynamic Hierarchy-based NSGA-II (Nondominated Sorting Genetic Algorithm II), called SDHN for short, to minimize both makespan and cost of workflow execution. First, a scoring criterion is developed to calculate the total score for each individual during population updating, which is used as a quantitative index to evaluate the dominance degree of individuals among the whole population. Hence, SDHN can distinguish individuals within the same dominance level and target its search toward the directions of elite solutions as their different dominance degrees and accordingly improve search efficiency. Second, a population-based dynamic hierarchical structure (HS) and its evolutionary rules are presented to update HS by comparing each child with all parental individuals from bottom to up until finding a proper dominant level. Since traversing all HS levels is not needed in most cases, the number of individual comparisons is reduced and SDHN&#x2019;s updating efficiency is greatly improved, especially for large-scale and complex applications. Third, to guarantee its converging to the near-optimal solutions, adaptive adjustment strategies (AASs) are designed to prevent the search from falling into local optima or diverging by checking the number of individuals at the highest HS level and then modifying the relevant genetic operations to guide the evolutionary process to approach the global Pareto Front. Extensive experiments are conducted to verify SDHN, and the results show that it outperforms the existing algorithms in the quality and diversity of resulting solutions as well as convergence time. <i>Note to Practitioners</i>&#x2014;Most scientific applications are computation and/or data-intensive and need large-scale or high-performance resources for their execution. More and more scientists use workflows to manage their applications, but how to efficiently run them in the cloud is a big challenge due to their large scale as well as the dynamic characteristics of the elastic and heterogeneous cloud resources. In this article, we develop a novel multiobjective optimization technique for workflow scheduling such that the makespan and cost can be minimized simultaneously. A scoring criterion, dynamic hierarchical structure and its evolutionary rules, and adaptive adjustment strategies are designed to cooperate with each other and increase the search ability and efficiency of the original and widely used NSGA-II. Adequate experiments are conducted to verify the proposed method&#x2019;s performance, and the experimental results show that it can provide more near-optimal solutions than the existing methods. It can be readily applied for implementing more efficient and effective cloud data centers to execute large-scale scientific workflows.

A Resource and Performance Optimization Reduction Circuit on FPGAs

Reduce is a fundamental computing pattern, which is widely involved in scientific and engineering applications. For example, accumulation, the most common example of reduce pattern, is the core of applications such as dot product, matrix multiplication, and finite impulse response (FIR) filter. However, there is a trade-off between performance and area in the hardware implementation of the reduce pattern. To solve this problem, we propose an optimized reduction method that can handle multiple arbitrary-length sets. The performance of the proposed method is evaluated for both a single data set and numerous data sets. Moreover, to quickly differentiate the data of different sets in the reduction circuit, individual modules are designed to manage the data. We implement the design on FPGAs and present the experimental results. The proposed design with high performance and low resource consumption can achieve at least 1.59 times improvement on area-time product compared with the reported methods.

Performance Prediction Method for Stream Computing Platform Based on Time Series

As one of the most popular high-performance data processing technology, existing task and resource scheduling strategies for stream computing platforms are suffering from the problem of triggering hysteresis, which seriously affects the cluster performance. To address this problem, the idea of performance prediction based on timeline series data is proposed. Firstly, the performance variation rule of the stream processing platform is analyzed, which provides a basis for proposing the performance prediction model. Secondly, the basic topology paradigm, periodic load prediction model, and real-time throughput prediction model are proposed as the theoretical foundation for performance prediction. Thirdly, the performance prediction algorithm is proposed to predict the variation trend of the processing load and throughput of the cluster. The processing load is predicted periodically while the throughput of the cluster is predicted in a real-time manner. Finally, the performance evaluation algorithm is proposed to evaluate the cluster performance based on the prediction results and trigger the corresponding scheduling strategies in advance. The experimental results showed that the prediction accuracy of the proposed algorithm meets the requirements in practical applications. Meanwhile, the proposed method improves the performance of stream computing performance by triggering the scheduling strategies in advance.

Enhancing Dynamic Binary Translation in Mobile Computing by Leveraging Polyhedral Optimization

Dynamic binary translation (DBT) is gaining importance in mobile computing. Mobile Edge Computing (MEC) augments mobile devices with powerful servers, whereas edge servers and smartphones are usually based on heterogeneous architecture. To leverage high‐performance resources on servers, code offloading is an ideal approach that relies on DBT. In addition, mobile devices equipped with multicore processors and GPU are becoming ubiquitous. Migrating x86_64 application binaries to mobile devices by using DBT can also make a contribution to providing various mobile applications, e.g., multimedia applications. However, the translation efficiency and overall performance of DBT for application migration are not satisfactory, because of runtime overhead and low quality of the translated code. Meanwhile, traditional DBT systems do not fully exploit the computational resources provided by multicore processors, especially when translating sequential guest applications. In this work, we focus on leveraging ubiquitous multicore processors to improve DBT performance by parallelizing sequential applications during translation. For that, we propose LLPEMU, a DBT framework that combines binary translation with polyhedral optimization. We investigate the obstacles of adapting existing polyhedral optimization in compilers to DBT and present a feasible method to overcome these issues. In addition, LLPEMU adopts static‐dynamic combination to ensure that sequential binaries are parallelized while incurring low runtime overhead. Our evaluation results show that LLPEMU outperforms QEMU significantly on the PolyBench benchmark.

Evolution of high-performance resources in the Institute of Automation and Control Processes FEB RAS

Evolution of high-performance resources in the Institute of Automation and Control Processes FEB RAS

Performance evaluation of a C-RAN supporting a mixture of random and quasi-random traffic

In this paper we consider a cloud radio access network (C-RAN) where the remote radio heads (RRHs) are separated from the baseband signal processing servers, named baseband units (BBUs). The latter forms a centralized pool of high-performance data center resources. To benefit from network function virtualization, we consider virtualized BBU (V-BBU) resources where the BBU functionality and services have been virtualized in the form of virtual network functions. All RRHs in the C-RAN form a single cluster. Each RRH of such a cluster may accommodate random or quasi-random traffic. That means that new calls in a RRH can be generated by an infinite number of mobile users (random traffic) or by a finite number of mobile users (quasi-random traffic). An arriving call requires a radio resource unit from the serving RRH and a computational resource unit from the V-BBU. If these resource units are available, then the call is accepted and remains in the system for a generally distributed service time. Otherwise, the call is blocked and lost. In order to analyze this C-RAN we model it as a loss system and study two cases: (i) all RRHs accommodate quasi-random traffic and (ii) some RRHs accommodate random traffic and the rest accommodate quasi-random traffic. In both cases, we show that the steady state probabilities have a product form solution and propose convolution algorithms for the accurate determination of the main teletraffic performance measure which is congestion probability. The accuracy of these algorithms is verified via simulation.

Delivering a machine learning course on HPC resources

In recent years, proficiency in data science and machine learning (ML) became one of the most requested skills for jobs in both industry and academy. Machine learning algorithms typically require large sets of data to train the models and extensive usage of computing resources, both for training and inference. Especially for deep learning algorithms, training performances can be dramatically improved by exploiting Graphical Processing Units (GPUs). The needed skill set for a data scientist is therefore extremely broad, and ranges from knowledge of ML models to distributed programming on heterogeneous resources. While most of the available training resources focus on ML algorithms and tools such as TensorFlow, we designed a course for doctoral students where model training is tightly coupled with underlying technologies that can be used to dynamically provision resources. Throughout the course, students have access to a dedicated cluster of computing nodes on local premises. A set of libraries and helper functions is provided to execute a parallelized ML task by automatically deploying a Spark driver and several Spark execution nodes as Docker containers. Task scheduling is managed by an orchestration layer (Kubernetes). This solution automates the delivery of the software stack required by a typical ML workflow and enables scalability by allowing the execution of ML tasks, including training, over commodity (i.e. CPUs) or high-performance (i.e. GPUs) resources distributed over different hosts across a network. The adaptation of the same model on OCCAM, the HPC facility at the University of Turin, is currently under development.

High Performance Fault Tolerant Resource Scheduling in Computational Grid Environment

Virtual resources team up to create a computational grid, which is used in computation-intensive problem solving. A majority of these problems require high performance resources to compute and generate results, making grid computation another type of high performance computing. The optimization in computational grids relates to resource utilization which in turn is achieved by the proper distribution of loads among participating resources. This research takes up an adaptive resource ranking approach, and improves the effectiveness of NDFS algorithm by scheduling jobs in those ranked resources, thereby increasing the number of job deadlines met and service quality agreements met. Moreover, resource failure is taken care of by introducing a partial backup approach. The benchmark codes of Fast Fourier Transform and Matrix Multiplication are executed in a real test bed of a computational grid, set up by Globus Toolkit 5.2 for the justification of propositions made in this article.

Interference-aware co-scheduling method based on classification of application characteristics from hardware performance counter using data mining

Computational scientists and engineers who are eager to obtain the best performance of scientific applications need efficient application characterization methods to successfully exploit high-performance hardware resources. However, modern processors are accompanied by high-bandwidth on-chip memory or a large number of cores. Therefore, application characterization research that takes into account the newly introduced hardware features in next-generation high performance computing environments is insufficient and complex. In this paper, we propose a simple and fast method to classify the application characteristics in systems state-of-the-art processors using hardware performance counters. The proposed method utilizes hardware performance counters to monitor hardware events related to system performance. A clustering approach is adopted that requires limited understanding of the correlation between hardware events and application characteristics. The application characterization technique is applied to NAS parallel benchmarks in two systems, including Intel Knights Landing and SkyLake Xeon processors. We demonstrate that the proposed techniques can capture system and application characteristics and provide users with useful insights into application execution.

Интернет-портал для решения научных и инженерных задач

Приведены описание, архитектура и сервисы инженерного вычислительного портала, представляющего собой комплексное решение для организации доступа к высокопроизводительным вычислительным ресурсам, кооперации специалистов из разных областей для решения сложных вычислительных задач, проведения вычислительных экспериментов и обучения параллельным технологиям. Goal. The project aimed at designing web-oriented high technology software and hardware complex (engineering and computational portal) to solve scientific and experimental problems which are carried out at Kemerovo State University. The project is considered to be unique, because three options, namely solution of engineering problems, performance of computational experiments based on high performance resources and educational services dedicated to theory and technology of high performance computations are combined. Methods. The architecture of the engineering and computational portal was designed with the help of a service-oriented method. This method is used to create distributed systems that have services for other systems or other services. The service-oriented method simplifies integration of new components to improve the portal’s options. The applied technology of web services is based on the following standards: WSDL - web service description standard, SOAP - web service interaction message format and BPEL - business processes description standard (orchestration), as well as portlet technology. The following tasks have already been accomplished: ∙ the web portal has been created on the basis of Liferay web platform; ∙ the service for user files storage management has been implemented; ∙ the service to solve flooded mine problem has been implemented; ∙ “Onlide” system has been developed in order to create software parallel user-code and start it on the basis of high performance resources; ∙ both components to register computational resources and to allow access to them have been implemented; ∙ “Virtual Laboratory Practicum” has been integrated as an educational element. Summary. The described engineering and computational portal is heavily employed at Kemerovo State University in order to perform computational experiments based on the high performance resources and to teach theory and technology of high performance computations. Future plans are as follows: implementation of the component to interact with the portal experts, “Master” component implementation, that is considered to be a portal guide, implementation of computational resources accounting components, implementation of the high-performance leased components, expansion of computational services list, implementation of the service to solve the problem of gas-dispersion gas/particle mixture combustion that applies Phoenics CFD technology.

Construing the big data based on taxonomy, analytics and approaches

Big data have become an important asset due to its immense power hidden in analytics. Every organization is inundated with colossal amount of data generated with high speed, requiring high-performance resources for storage and processing, special skills and technologies to get value out of it. Sources of big data may be either internal or external to organization, and big data may reside in structured, semi-structured or unstructured form. Artificial intelligence, Internet of Things, and social media are contributing to the growth of big data. Analytics is the use of statistics, maths, and machine learning to derive meaningful insights from data to make timely decisions and enable data-driven organization of the future. This paper sheds light upon big data, taxonomy of data, and hierarchical journey of data from its original form to the high level understanding in terms of wisdom. The paper also focuses on key characteristics of big data and challenges of handling big data. In addition, big data storage systems have also been briefly covered to get the idea on how storage systems help to accommodate the requirements of big data. This paper scrupulously articulates the eras of evolution of analytics varying from descriptive, predictive and prescriptive analytics. Process models used for inferring information from data have been compared and their applicability for analyzing big data has also been sought. Finally, recent developments carried in the domain of big data and analytics are compared based on the state-of-the-art approaches.

Ergodic optimization of stochastic differential systems in wireless networks

The present work is the second article in a couple of intertwined papers. They form complementary items on the same subject. They both address the problem of joint power allocation and time slot scheduling in a wireless communication system with time varying traffic. The system is handled by a single base station transmitting over time varying channels. The operating time horizon is divided into time slots, and a fixed amount of power is available at each time slot. The mobile users share each time slot and the power available at this time slot. Since many wireless network applications have stringent delay requirements, designing high-performance resource allocation algorithms to achieve minimum possible delay is of great importance, and this is the main objective of the work presented in this paper. The delay performance of a resource allocation algorithm can be characterized by the average delay experienced by the data transmitted in the network. We propose a heavy traffic analysis for the physical system on hand, i.e., appropriate re-scaling that leads to a diffusion approximation of the system in the sense of weak convergence. The approximate diffusion is constrained or bounded in the K-dimensional positive orthant. We establish the convergence result of the heavy traffic analysis, and then a closed form solution to the resource optimization problem is provided. Here the solution relies on the ergodicity of the approximate diffusion.

Analysis of Dynamic Virtual Private Networks Resource Allocation Schemes

The need for high performance resource allocation schemes for Virtual Private Networks (VPNs) has led to the proliferation of algorithms for VPN resource allocation. It was found that most works on VPN resource allocation focused either on admission control or link reservation on the network. Also, review of relevant literatures has revealed the need for a resource allocation/scheduling scheme whose algorithm will be able to allocate bandwidth and memory resources to different VPNS sharing the same link to the network service provider. Resources should be allocated in such a manner that utilization is optimal while VPN endpoints or customers receive services that do not undermine the service-level agreement (SLA) with the service provider. MATLAB Simulink was used to design a simulation model for analysing the VPN access network obtaining and comparing results for the link bandwidth utilization, buffer memory utilization and packet loss rate performances of the RDVNP (Robust Dynamic Virtual Network Provisioning) algorithm against the DWARF-Net (Dynamic bandwidth Allocation and guarantee on Resource Fairness) algorithm. From the results obtained, DWARF-Net algorithm’s performance was better than the RDVNP’s algorithm in almost all parameters tested on. On bandwidth utilization, DWARF-Net had an average channel utilization of 61.23% against RDVNP’s 48.28%, on buffer utilization 42% for DWARF-Net, 41% for RDVNP and average loss rate average of 1 Packet/second for DWARF-Net against 20 Packets/second for RDVNP. From the simulation analysis and result of this work, DWARF-Net is recommended as an optimal performing algorithm for VPN resource allocation.

DSM: a dynamic scheduling method for concurrent workflows in cloud environment

Cloud computing, emerged as a commercial service model, has been widely concerned in both industry and academia. Massive workflow applications could be performed simultaneously on the cloud platforms, which significantly benefits from the elasticity and convenience of cloud computing. However, it is still a challenge to schedule virtualized resources for the concurrent workflows in cloud environment, with limited high-performance resources in a timesaving and efficient manner. In view of this challenge, a dynamic scheduling method for concurrent workflows, named as DSM, in cloud environment is proposed to satisfy the various resource requirements of the workflows. Technically, a time overhead model for the workflows and a resource utilization model for cloud datacenter are presented. Then a relevant dynamic scheduling method is designed based on critical path lookup, which aims at minimizing the makespan of workflows, and maximizing the resource utilization of the datacenter during the execution of the workflows. Extensive experimental evaluations demonstrate the efficiency and effectiveness of our proposed method.