Distribution Of Computational Load Research Articles

A dynamic load balancing model coupled with DAC and ISAT for a stochastic turbulent combustion model

Due to the composition-dependent stiffness of chemistry, simulations of reactive turbulent flows may present computational load imbalance among parallel processes when spatial decomposition is used for parallelisation, causing high CPU idle time and waste of computational resources. To increase computational efficiency, a dynamic load balancing (DLB) model is proposed to redistribute computational load among computing cores. The DLB model exploits a decomposition in the mixture fraction space with two dynamic adjusting decomposition strategies to realise load redistribution. The DLB model is suitable for the integration of chemistry on stochastic particles in hybrid Eulerian/Lagrangian turbulent combustion models in which the Eulerian field is conventionally decomposed statically in physical space in a way that balances the computational load for the solution of the Navier-Stokes equation but which does not generally lead to balanced load for the computation of the composition fields. Here it is tested using an OpenFOAM-based platform, mmcFoam, which is a comprehensive object-orientated C++ library for stochastic turbulent combustion modelling. Apart from direct integration (DI) for chemistry, the DLB model is also coupled with dynamic adaptive chemistry (DAC) and in situ adaptive tabulation (ISAT), which allows for extra speedup. The performance of the coupled models is validated and assessed for two laboratory flame conditions that exhibit different levels of computational load imbalance. Overall, the DLB model effectively balances the computational load distribution and increases the effective usage of computing power, shortening the simulation wall time required. Moreover, a strong scaling test is carried out using up to 512 cores. Although all approaches have sub-ideal scalability, the scalability of each with DLB is significantly better than without DLB. While DLB-ISAT has relatively poor scalability compared to the DI- and DAC-based methods, DLB-ISAT still ranks the fastest among the algorithms in all scaling trials.

FedDOVe: A Federated Deep Q-learning-based Offloading for Vehicular fog computing

Connected Autonomous Vehicles (CAVs) aim to provide various smart transportation applications which have computation-intensive tasks. The vehicles having less availability of computational resources offload tasks to roadside units (RSUs) for processing. Renewable energy-powered RSUs have limited energy, storage, and computational capabilities. To reduce the computation load at RSUs, vehicular fog computing is used for computation-intensive tasks. Vehicular fog computing brings the computational resources nearer to the requesting devices for reducing the latency in the request fulfillment. RSU computes the tasks either locally or offloads computation-intensive tasks to the fog server for further processing. In vehicular fog computing, one of the most important problems is to find an optimal association of RSUs with fog servers for computation offloading. A significant energy consumption occurs across RSUs in offloading computation-intensive tasks to fog servers. On the other hand, fog servers may have uneven computation load due to varying offloading rate across the road segments. The non-uniform distribution of computation load across fog servers may increase the latency in the request fulfillment. In this paper, we propose an online Federated Deep Q-learning-based Offloading technique for Vehicular fog computing (FedDOVe) which jointly optimizes the energy consumption across RSUs and load balancing across fog servers. FedDOVe is a model-free reinforcement learning approach that uses the global information for finding an optimal association between RSUs and fog servers. Simulation results show that the proposed FedDOVe reduces the energy consumption by about 47%–49% and improves load balancing by about 60%–65% as compared to other existing offloading techniques.

A design framework for a system of digital twins and services

Digital twins represent physical systems in virtual space to enable knowledge intensive tasks. For complex physical systems, effective digital integration can often be achieved by means of a system of aggregated digital twins coupled with a services network. Designing such a system of digital twins and services involves a daunting array of options and considerations. This paper presents a design framework that facilitates systematic, effective decisions when designing a system of digital twins to integrate the data from a complex physical system. The design framework is arranged in six steps: 1) needs and constraints analysis, 2) physical system decomposition, 3) services allocation, 4) performance and quality considerations, 5) implementation considerations and 6) verification and validation. The design framework works with a general reference architecture that combines a digital twin aggregation hierarchy with a service-oriented architecture. This reference architecture allows for the separation of concerns, computational load distribution, incremental development and modular software design. The paper considers the merits of the design framework using two illustrative case studies, i.e. a heliostat field and a smart city. The case studies indicate that the design framework can be applied to a wide range of complex systems and can improve the effectiveness of the design process.

ANumerical Uncertainty in Parallel Processing Using Computational Fluid Dynamics as Example

Computational methods such as Computational Fluid Dynamics (CFD) traditionally yield a single output – a single number that is much like the result one would get if one were to perform a theoretical hand calculation. However, this paper will show that computation methods have inherent uncertainty which can also be reported statistically. In numerical computation, because many factors affect the data collected, the data can be quoted in terms of standard deviations (error bars) along with a mean value to make data comparison meaningful. In cases where two data sets are obscured by uncertainty, the two data sets are said to be indistinguishable. A sample CFD problem pertaining to external aerodynamics is copied and ran on 29 identical computers in a university computer lab. The expectation is that all 29 runs should return exactly the same result; unfortunately, in a few cases the result turns out to be different. This is attributed to the parallelization scheme which partitions the mesh to run in parallel on multiple cores of the computer. The distribution of the computational load is hardware-driven depending on the available resource of each computer at the time. Things, such as load-balancing among multiple Central Processing Unit (CPU) cores using Message Passing Interface (MPI) are transparent to the user. Software algorithm such as METIS or JOSTLE is used to automatically divide up the load between different processors. As such, the user has no control over the outcome of the CFD calculation even when the same problem is computed. Because of this, numerical uncertainty arises from parallel (multicore) computing. One way to resolve this issue is to compute problems using a single core, without mesh repartitioning. However, as this paper demonstrates even this is not straight forward. Keywords: numerical uncertainty, parallelization, load-balancing, automotive aerodynamics

Effects of Differentiated 5G Services on Computational and Radio Resource Allocation Performance

5G is poised to support new emerging service types that help in the realization of futuristic applications. These services include enhanced Mobile BroadBand (eMBB), ultra-Reliable Low Latency Communication (uRLLC), and massive Machine-Type Communication (mMTC). Even though the new services offer a variety of new use-cases to be implemented, it is still a challenge to guarantee the Quality of Service (QoS) they demand. Moreover, as considerable amount of computational resources are introduced in the evolved Radio Access Network (RAN) following the Mobile Edge Computing (MEC) concept, computational resource allocation optimization along with radio allocation becomes essential. In this paper, we examine the characteristics of the new 5G services and propose a joint computational and radio resource allocation framework that analyzes the QoS performance of each 5G service individually. The framework is developed based on per-service load characterization. Therefore, a computational load distribution algorithm is developed that balances the workloads subject to user association constraint. Further, radio resource allocation performs load-based eMBB-mMTC slicing and uRLLC puncturing. The simulation results show that the proposed solution reduces the packet loss ratio by up to 15% and increases the user data rate by up to 7% for 4G-like services. Furthermore, the effect of resource granularity in radio allocation has been identified as crucial factor for effective allocation of services with small data loads. Finally, the problem of small granularity has been solved by adapting the allocation interval.

Reliable distribution of computational load in robot teams

Modern multi-robot systems often need to solve computationally intensive tasks but operate with limited compute resources and in the presence of failures. Cooperating to share computational tasks between robots at the edge reduces execution time. We introduce and evaluate a new computation load management technology for teams of robots: Reliable Autonomous Mobile Programs (RAMPs). RAMPs use information about the computational resources available in the team and a cost model to decide where to execute. RAMPs are implemented in ROS on a collection of Raspberry Pi-based robots. The performance of RAMPs is evaluated using route planning, a typical computationally-intensive robotics application. A systematic study of RAMPs demonstrates a high likelihood of optimal or near-optimal distribution and hence efficient resource utilisation. RAMPs successfully complete in the presence of simultaneous, or successive, robot failures and network failures, while preserving near-optimal distribution.

Analysis of training of deep neural networks with heterogeneous architecture while detecting malicious network traffic

The given article presents the study of the training process of a composite heterogeneous neural network with deep architecture. A brief description of the architecture of the analyzed neural network system has been given. The key nodes of computational load distribution on each of the layers of the neural network have been tracked. The monitoring results have been analyzed and the following conclusions have been made: the most resource-intensive part of the system during training is the LSTM network. The results of the study of structural features of the neural network hidden layers have been presented. Graphs have been constructed and there has been carried out the study of a statistical distribution of weights on each unique layer of the architecture: on a fully connected layer, on the first hidden layer of the subnet-encoder, on the second hidden layer of the subnet-encoder, on the first and second fully connected output layers of the neural network. Based on the research results, a qualitative assessment of the effectiveness and accuracy of the entire neural network system has been given.

Load balancing in data distribution systems

The issue of distribution of large amounts of information on server stations is researched. The analysis of existing methods for distributing data of heavily loaded web applications is carried out. A system for computing load distribution using a balancing server is proposed. A description of the software necessary for the functioning of the system and server management interface is provided.

A Complexity Analysis of Event-Triggered Model Predictive Control on Industrial Hardware

We implement a recently proposed event-triggered networked model predictive control (MPC) approach on industrial hardware to analyze its practical relevance. There exist several alternatives for such implementation which differ with respect to the distribution of computational load between local and central nodes and with respect to network bandwidth requirements. These alternatives have been analyzed theoretically before, but, when implemented, it becomes evident that their usefulness cannot be predicted based on theoretical considerations alone. It is the purpose of this brief to account for both practical and theoretical aspects in determining which alternative is most appropriate for implementation on industrial hardware. The smallest possible bandwidth is known to result for a variant in which only the active set of constraints is transmitted from the central to the local nodes. Since local nodes must determine the control law from the active set in this case, which requires matrix inversions, an unattractive computational cost results at first sight. Somewhat surprisingly, the computational cost scales practically linearly in the problem size when implemented. We confirm this result with a more detailed theoretical complexity analysis than given in previous articles. All results are illustrated with data obtained with implementation of industrial hardware components.

Modeling the parallelism of empirical models of optimal complexity using a Petri net

Many physical processes and phenomena in view of their complexity cannot be described analytically. In these cases, empirical modeling is applied. In this research, the method based on the genetic approach is used to construct empirical models of optimal complexity that have the form of a polynomial of assigned power. Implementation of the developed method requires a multiple solution of the system of linear algebraic equations. Solution of the system of linear algebraic equations is found by reducing the corresponding matrix to the upper diagonal form with unities on the main diagonal. Analysis of the algorithm showed that the procedure of reducing the matrix to the upper diagonal form has internal parallelism. Based on the created model of the computational process in the form of a Petri net, the strategy of construction of the parallel algorithm for solving the system of linear algebraic equations was developed. The essence of the strategy is that computations are performed on some parallel processors. One of them was assigned coordinating functions, and it was named master. Other processors – slaves are subordinated to the master. Division of computation volume is such that the number of rows of the matrix, which master operates is at least by unity more than the corresponding number of rows allocated to the servant. The effectiveness of the parallel algorithm for the proposed strategy was evaluated based of the criterion of the total number of arithmetic operations. The proposed strategy is an integral part of the process of synthesis of the empirical model of optimal complexity based on the genetic algorithms. Distribution of computational load between processors working in parallel (master and slaves) ensures the acceleration of the computational process by five times or more.

РАСПРЕДЕЛЕНИЕ ДАННЫХ В ИНФОРМАЦИОННОЙ СИСТЕМЕ С ПОМОЩЬЮ СЕРВЕРА-БАЛАНСИРА

The issue of distribution of large amounts of information system information on server stations is considered. The analysis of existing methods for distributing data of highly loaded web applications is carried out. A system for computing load distribution using a balancing server is proposed. A description of the software necessary for the functioning of the system, and a server management interface are provided.

Multiagent system for dynamic resource provisioning in cloud computing platforms

Dynamic provisioning of computational resources in cloud platforms is a challenging problem. In this scenario, multiagent system technology may offer noticeable improvements, with intelligent agents to dynamically choose the best computational load distribution reducing execution time and cloud services cost. Thus, this article presents MAS-Cloud, a multiagent system to dynamically monitor, predict and provide computational resources in cloud platforms. Deductive reasoning agents work cooperatively in a three layer architecture to provide transparent horizontal elasticity of virtual machines in public cloud platforms (i.e., Google, Amazon EC2). MAS-Cloud was evaluated with a nondeterministic and CPU intensive application providing a challenging validation case. A multiple linear regression model was used presenting good predictions with reduced mean absolute percent error achieving 3.59% CPU usage and 5.28% of execution time in the Google platform, and 3.97% CPU usage and 5.47% of execution time in the Amazon EC2 platform.

A Framework to Design the Computational Load Distribution of Wireless Sensor Networks in Power Consumption Constrained Environments.

In this paper, we present a work based on the computational load distribution among the homogeneous nodes and the Hub/Sink of Wireless Sensor Networks (WSNs). The main contribution of the paper is an early decision support framework helping WSN designers to take decisions about computational load distribution for those WSNs where power consumption is a key issue (when we refer to “framework” in this work, we are considering it as a support tool to make decisions where the executive judgment can be included along with the set of mathematical tools of the WSN designer; this work shows the need to include the load distribution as an integral component of the WSN system for making early decisions regarding energy consumption). The framework takes advantage of the idea that balancing sensors nodes and Hub/Sink computational load can lead to improved energy consumption for the whole or at least the battery-powered nodes of the WSN. The approach is not trivial and it takes into account related issues such as the required data distribution, nodes, and Hub/Sink connectivity and availability due to their connectivity features and duty-cycle. For a practical demonstration, the proposed framework is applied to an agriculture case study, a sector very relevant in our region. In this kind of rural context, distances, low costs due to vegetable selling prices and the lack of continuous power supplies may lead to viable or inviable sensing solutions for the farmers. The proposed framework systematize and facilitates WSN designers the required complex calculations taking into account the most relevant variables regarding power consumption, avoiding full/partial/prototype implementations, and measurements of different computational load distribution potential solutions for a specific WSN.

Decomposition strategies in the problems of simulation of additive laser technology processes

The development of additive technologies and their application in industry is associated with the possibility of predicting the final properties of a crystallized added material. This paper describes the problem characterized by a dynamic and spatially nonuniform computational complexity, which, in the case of uniform decomposition of a computational domain, leads to an unbalanced load on computational cores. The strategy of partitioning of the computational domain is used, which minimizes the CPU time losses in the serial computations of the additive technological process. The chosen strategy is optimal from the standpoint of a priori unknown dynamic computational load distribution. The scaling of the computational problem on the cluster of the Institute on Laser and Information Technologies (RAS) that uses the InfiniBand interconnect is determined. The use of the parallel code with optimal decomposition made it possible to significantly reduce the computational time (down to several hours), which is important in the context of development of the software package for support of engineering activity in the field of additive technology.

A low-level resource allocation in an agent-based Cloud Computing platform

The distribution of computational resources in a Cloud Computing platform is a complex process with several parameters to consider such as the demand for services, available computational resources and service level agreements with end users. Currently, the state-of-the-art presents centralized approaches derived from previous technologies related to cluster of servers. These approaches allocate computational resources by means of the addition/removal of (physical/virtual) computational nodes. However, virtualization technology currently allows for research into new techniques, which makes it possible to allocate at a lower level. In other words, not only is it possible to add/remove nodes, but also to modify the resources of each virtual machine (low level resource allocation). Thus, agent theory is a key technology in this field, allowing decentralized resource allocation. This innovative approach has undeniable improvements such us computational load distribution and reduced computation time. The evaluation was carried out through experiments in a real Cloud environment, thus proving the validity of the proposed approach.

The Method of Constructing an Integrated Corporate Information System

Background/Objectives: This article describes the integration method of mobile applications with Corporate Information System (CIS) through the previously proposed interaction architecture based on integration environment. Methods: The methods of data obtaining and processing and computational load distribution between the components of an integrated corporate system were used to develop a method of mobile application integration with corporate information system, implemented as an algorithm of building an integration environment for cross-platform mobile applications. Findings: Within the framework of this article, the method implementing the process of developing mobile application integrated with the CIS in accordance with the proposed architecture was discussed. The proposed method involves the use of different types of models: transforming data model, data storage model and data representation model. These models combine all necessary technologies of procurement, processing and representation of data circulating in the CIS. The structure of the integration environment, which includes new approaches to the organization of data storage, conversion and presentation, is described in detail. Also this article represents the algorithm for implementation of CIS mobile applications integration including the proposed structure of integration environment. Applications/Improvements: The proposed technology can be used to build various integrated solutions for the education sphere, transportation, security, provided the simultaneous development of mobile applications for multiple mobile platforms, or at the integration of several CIS.

High-performance programming frameworks for numerical simulation

Over the past 20 years, a variety of research activities on high-performance programming have taken place worldwide. General-purpose application programming interfaces (GAPIs) have been designed and standardized; for example, MPI and OpenMP are two successfulGAPIs in the field of numerical simulation. However, with the increase in performance of computer systems from teraflops to petaflops [1], these GAPIs will not be adequate to deal with emerging systems in the next 5–10 years. There are two main reasons for this [2]. First, GAPIs need to be significantly improved or redesigned to deal with the multilevel heterogeneous parallelism or deep memory hierarchies mandated by the power and microarchitecture considerations. Second, domain-specific abstractions for application programming interfaces (DAPIs) are essential to shield application developers from the complexity and availability of such GAPIs. To date, high-performance programming frameworks (HPPFs) have successfully been used in implementing DAPIs in the field of numerical simulation [2]. These frameworks can bridge the gap between application developers and GAPIs with the paradigm of ‘think in parallel and write sequentially’. Physics and numerical algorithms can be programmed sequentially based on a flowchart of the parallel code without any knowledge of GAPIs. Moreover, HPPFs allow parallel code to be ported from petaflop systems to emerging or exascale systems without having to be rewritten. In other words, HPPFs safeguard application software from changes due to evolving computer systems. Typically, HPPFs are in greater demand in institutes where there are more application domain programmers than high-performance programmers. Figure 1 depicts the inner workings of a HPPF. First, domain-specific data structures must be presented for the distribution of computation load toward potentially minimal data movement across thememory hierarchy for efficient implementation of numerical algorithms.Next, the domain-specific data dependencies of the numerical algorithms must be accurately described for the new data structures. Third, parallel computation models (PCMs) can be defined to depict different types of numerical computation phases. Here, each phase represents a type of data dependency among a set of sequential numerical computation entities. Fourth, parallel algorithms for data communication and load balancing must be designed for efficient implementation of the PCMs. Finally, DAPIs are defined as software components [3] using C++ constructs in which data structures are available and numerical computation entities are integrated. Then, the numerical algorithms can be implemented sequentially provided that they can be written as a flowchart comprising numerical computation phases for each existing type of PCM. Further, numerical algorithms can be classified into two types, namely, application aware and independent. The

Computational Load Balancing Algorithm for Parallel Knapsack Packing Tree Traversal

The paper considers efficient computational load distribution for the exact parallel algorithm for the knapsack problem based on packing tree search. We propose an algorithm that provides for static and dynamic computational load balancing for the problem in question.

Maintenance of sustainable operation of pipeline-parallel computing systems in the cloud environment

This paper reviews the model and algorithm for the intelligent scheduling of a pipelineparallel computing process, which takes into account changes in the resource consumption of computing nodes. The solution is implemented in a virtual cloud environment, which provides security and accessibility of high-performance computing services due to the isolation of physical resources and the monitoring and optimization of their use, as well as the adaptive distribution of computational load.

Runtime balancing effect in distributed simulation model

The paper deals with study of the load distribution effect between computing nodes for simulation run-time of a distributed computer network model. Two main types of balancing are considered: computational load balancing and reduction of transmitted amounts of data. Simulation was performed on one computer, and distribution was carried out between the cores of one processor. Simulation experiments showed that the lack of balance in the amounts of data, transferred between parts of a distributed model, leads to decrease of the simulation speed by several times due to overhead charges for data transmission between logical processes because of MPI usage. Model run-time changing at uneven distribution of computational load depends to a large extent on the load, which is created by the applications running on the simulated network nodes. It is shown that a balanced model is performed much faster than unbalanced version even when using applications that do not require significant computing resources. Simulation time reduction can be achieved by model separation in such manner as to reduce amounts of data transferred between its parts and reduce variability of loads generated by applications in different logical processes.