Message Passing Interface Protocol Research Articles

Comparison of edge computing methods in Internet of Things architectures for efficient estimation of indoor environmental parameters with Machine Learning

The large increase in the number of Internet of Things (IoT) devices have revolutionised the way data is processed, which added to the current trend from cloud to edge computing has resulted in the need for efficient and reliable data processing near the data sources using energy-efficient devices. Two methods based on low-cost edge-IoT architectures are proposed to implement lightweight Machine Learning (ML) models that estimate indoor environmental quality (IEQ) parameters, such as Artificial Neural Networks of Multilayer Perceptron type. Their implementation is based on centralised and distributed parallel IoT architectures, connected via wireless, which share commercial off-the-self modules for data acquisition and sensing, such as sensors for temperature, humidity, illuminance, CO2, and other gases. The centralised method uses a Graphics Processing Unit and the Message Queuing Telemetry Transport protocol, but the distributed method utilises low-performance ARM-based devices and the Message Passing Interface protocol. Although multiple IEQ parameters are measured, the training and testing of ML models is accomplished with experiments focused on small temperature and illuminance datasets to reduce data processing load, obtained from sudden spikes, square profiles and sawteeth test cases. The results show a high estimation performance with F-score and Accuracy values close to 0.95, and an almost theorical Speedup with a reduction in power consumption close to 37% in the distributed parallel approach. In addition, similar or slightly better performance is achieved compared to equivalent IoT architectures from related research, but error reduction of 35–76% is accomplished with an adequate balance between performance and energy efficiency.

Hydrodynamic Coupling of Viscous and Nonviscous Numerical Wave Solutions Within the Open-Source Hydrodynamics Framework reef3d

Abstract A comprehensive understanding of the marine environment in the offshore area requires phase-resolved wave information. For far-field wave propagation, computational efficiency is crucial, as large spatial and temporal scales are involved. For the near-field extreme wave events and wave impacts, high resolution is required to resolve the flow details and turbulence. The combined use of a computationally efficient large-scale model and a high-resolution local-scale solver provides a solution that combines accuracy and efficiency. This article introduces a coupling strategy between the efficient fully nonlinear potential flow (FNPF) solver REEF3D::FNPF and the high-fidelity computational fluid dynamics (CFD) model REEF3D::CFD within the open-source hydrodynamics framework REEF3D. REEF3D::FNPF solves the Laplace equation together with the boundary conditions on a sigma-coordinate. The free surface boundary conditions are discretized using high-order finite difference methods. The Laplace equation for the velocity potential is solved with a conjugated gradient solver preconditioned with a geometric multigrid provided by the open-source library Hypre. The model is fully parallelized following the domain decomposition strategy and the message passing interface protocol. The waves calculated with the FNPF solver are used as wave generation boundary conditions for the CFD-based numerical wave tank REEF3D::CFD. The CFD model employs an interface capturing two-phase flow approach that can resolve complex wave structure interaction, including breaking wave kinematics and turbulent effects. The presented hydrodynamic coupling strategy is tested for various wave conditions and the accuracy is fully assessed.

The Data Acquisition System of the KOTO Experiment and the RCE Platform Technology Upgrade

The KOTO experiment at J-PARC, Japan, aims to observe the rare neutral kaon decay mode $K_{L}\rightarrow \pi ^{0}\nu \bar {\nu }$ . After the first experimental run in May 2013 at a 24-kW beam power, the KOTO data acquisition (DAQ) system was upgraded in 2015 to provide efficient and reliable data collection at higher beam intensities. Lossless data compression in the analog-to-digital converter modules was implemented to reduce the size of data packets, resulting in a threefold increase in data collection rate. A new software trigger on a 47-node cluster was designed to use Infiniband hardware with message passing interface protocol to establish a mesh network inside the computer clusters for parallel data processing. The upgrade to the KOTO DAQ system was commissioned in 2015 and successfully collected data with a beam intensity of up to 42 kW. In preparation for increasing beam intensities in future runs, the hardware trigger upgrades using the reconfigurable clustering element platform technology are under development.

Parallel finite element analysis of seismic soil structure interaction using a PC cluster

This paper describes an implementation of a highly scalable parallel computational facility with high speedup efficiency using relatively low-cost hardware, which consists of a cluster of desktop personal computers (PCs) connected via a 10-Gigabit Ethernet. Two-levels of parallelization were implemented. Communication between different PCs was achieved using message passing interface (MPI) protocol. Domain decomposition was automated and based on element numbering. Domain continuity was assured largely by re-numbering the elements using a “front squasher” code prior to decomposition. Within each PC, the shared memory parallelization was implemented using either the open-multiprocessing (OpenMP) or the MPI protocol. Analysis of three different problems with number of degrees-of-freedom ranging from about 129,000 to about 2,260,000 shows a speedup efficiency generally above 70%. Super-linear speedup was achieved in several of the cases examined in this study, with the hybrid MPI-OpenMP approach generally performing better compared to the pure MPI method for parallelization. The results demonstrate the feasibility of acquiring a parallel computing facility with relatively modest outlay that is within the reach of consulting or engineering offices.

New Simulation Schemes and Capabilities for the PHISICS/RELAP5-3D Coupled Suite

PHISICS (Parallel and Highly Innovative Simulation for INL Code System) is a reactor physics package developed at the Idaho National Laboratory. It is composed of several modules: a nodal and semistructured transport core solver (INSTANT), a depletion module (MRTAU), a time-dependent solver (TimeIntegrator), a cross-section interpolation and manipulation framework (MIXER), a criticality search module (CRITICALITY), and a fuel management and shuffling component (SHUFFLE). The PHISICS code has been coupled to the RELAP5-3D thermal-hydraulics code. Flexibility in the coupling among the different modules and with RELAP5-3D allows for several new integrated computational schemes and improvements with respect to current available options using NESTLE/RELAP5-3D. These schemes will be described in this paper. Moreover, the whole PHISICS package is fully parallelized, using the Message Passing Interface protocol. This allows for reduced computational times, while providing the capability to solve very detailed problems.

Parallel Computing for Sorting Algorithms

The expanding use of multi-processor supercomputers has made a significant impact on the speed and size of many problems. The adaptation of standard Message Passing Interface protocol (MPI) has enabled programmers to write portable and efficient codes across a wide variety of parallel architectures. Sorting is one of the most common operations performed by a computer. Because sorted data are easier to manipulate than randomly ordered data, many algorithms require sorted data. Sorting is of additional importance to parallel computing because of its close relation to the task of routing data among processes, which is an essential part of many parallel algorithms. In this paper, sequential sorting algorithms, the parallel implementation of many sorting methods in a variety of ways using MPICH.NT.1.2.3 library under C++ programming language and comparisons between the parallel and sequential implementations are presented. Then, these methods are used in the image processing field. It have been built a median filter based on these submitted algorithms. As the parallel platform is unavailable, the time is computed in terms of a number of computations steps and communications steps

Parallel Computing for Sorting Algorithms

The expanding use of multi-processor supercomputers has made a significant impact on the speed and size of many problems. The adaptation of standard Message Passing Interface protocol (MPI) has enabled programmers to write portable and efficient codes across a wide variety of parallel architectures. Sorting is one of the most common operations performed by a computer. Because sorted data are easier to manipulate than randomly ordered data, many algorithms require sorted data. Sorting is of additional importance to parallel computing because of its close relation to the task of routing data among processes, which is an essential part of many parallel algorithms. In this paper, sequential sorting algorithms, the parallel implementation of many sorting methods in a variety of ways using MPICH.NT.1.2.3 library under C++ programming language and comparisons between the parallel and sequential implementations are presented. Then, these methods are used in the image processing field. It have been built a median filter based on these submitted algorithms. As the parallel platform is unavailable, the time is computed in terms of a number of computations steps and communications steps

Self-Consistent MPI Performance Guidelines

Message passing using the Message-Passing Interface (MPI) is at present the most widely adopted framework for programming parallel applications for distributed memory and clustered parallel systems. For reasons of (universal) implementability, the MPI standard does not state any specific performance guarantees, but users expect MPI implementations to deliver good and consistent performance in the sense of efficient utilization of the underlying parallel (communication) system. For performance portability reasons, users also naturally desire communication optimizations performed on one parallel platform with one MPI implementation to be preserved when switching to another MPI implementation on another platform. We address the problem of ensuring performance consistency and portability by formulating performance guidelines and conditions that are desirable for good MPI implementations to fulfill. Instead of prescribing a specific performance model (which may be realistic on some systems, under some MPI protocol and algorithm assumptions, etc.), we formulate these guidelines by relating the performance of various aspects of the semantically strongly interrelated MPI standard to each other. Common-sense expectations, for instance, suggest that no MPI function should perform worse than a combination of other MPI functions that implement the same functionality, no specialized function should perform worse than a more general function that can implement the same functionality, no function with weak semantic guarantees should perform worse than a similar function with stronger semantics, and so on. Such guidelines may enable implementers to provide higher quality MPI implementations, minimize performance surprises, and eliminate the need for users to make special, nonportable optimizations by hand. We introduce and semiformalize the concept of self-consistent performance guidelines for MPI, and provide a (nonexhaustive) set of such guidelines in a form that could be automatically verified by benchmarks and experiment management tools. We present experimental results that show cases where guidelines are not satisfied in common MPI implementations, thereby indicating room for improvement in today's MPI implementations.

General Subdomain Boundary Mapping Procedure for Structured Grid Implicit CFD Parallel Computation

In the present work, a general subdomain boundary mapping procedure has been developed for arbitrary topology multiblock structured grids with grid points matched on subdomain boundaries. The interface of two adjacent blocks is uniquely defined according to each local mesh index system, which is specified independently.A pack/unpack procedure basedonthedefinitionoftheinterfaceisdevelopedtoexchangethedatainaone-dimensional array to minimize data communication. A secure send/receive procedure is employed to remove the possibility of blocked communication and achieve optimum parallel computationefficiency.Twoterms,“order”and“orientation”,areintroducedasthelogicsdefiningthe relationship of adjacent blocks. The procedure is applied to parallelize a three-dimensional Navier‐Stokes code, which uses an implicit time marching scheme with line Gauss‐Seidel iteration. The partitioning of the implicit matrix is done by discarding the corner matrices, which is easily implemented and is shown to have a small negative effect on the convergence rate. The message passing interface protocol is used for communicating the data. The numerical experiments presented in this paper include two- and three-dimensional flows using Reynolds averaged Navier‐Stokes equations and a detached eddy simulation with a fifth-orderweightedessentiallynon-oscillatoryscheme.Numericalexperimentsonamessage passing interface based computer cluster show that this general mapping algorithm is robust and has high parallel computing efficiency.

Accelerated 3D-OSEM image reconstruction using a Beowulf PC cluster for pinhole SPECT

A conventional pinhole single-photon emission computed tomography (SPECT) with a single circular orbit has limitations associated with non-uniform spatial resolution or axial blurring. Recently, we demonstrated that three-dimensional (3D) images with uniform spatial resolution and no blurring can be obtained by complete data acquired using two-circular orbit, combined with the 3D ordered subsets expectation maximization (OSEM) reconstruction method. However, a long computation time is required to obtain the reconstruction image, because of the fact that 3D-OSEM is an iterative method and two-orbit acquisition doubles the size of the projection data. To reduce the long reconstruction time, we parallelized the two-orbit pinhole 3D-OSEM reconstruction process by using a Beowulf personal computer (PC) cluster. The Beowulf PC cluster consists of seven PCs connected to Gbit Ethernet switches. Message passing interface protocol was utilized for parallelizing the reconstruction process. The projection data in a subset are distributed to each PC. The partial image forward- and back-projected in each PC is transferred to all PCs. The current image estimate on each PC is updated after summing the partial images. The performance of parallelization on the PC cluster was evaluated using two independent projection data sets acquired by a pinhole SPECT system with two different circular orbits. Parallelization using the PC cluster improved the reconstruction time with increasing number of PCs. The reconstruction time of 54 min by the single PC was decreased to 10 min when six or seven PCs were used. The speed-up factor was 5.4. The reconstruction image by the PC cluster was virtually identical with that by the single PC. Parallelization of 3D-OSEM reconstruction for pinhole SPECT using the PC cluster can significantly reduce the computation time, whereas its implementation is simple and inexpensive.

THE EFFECTS OF CLUSTER SIZE ON PACKET INTER-ARRIVAL PATTERNS AND INTENSITY IN A DISTRIBUTED SYSTEM

Although the design and implementation of parallel processing infrastructure is not considered a mainstream IS topic, a working knowledge of its capabilities can have an important influence on improving the design of on-line systems. In order to illustrate some of the advantages and limitations of using parallel processing in information systems two experiments were run to collect multiple node scaling information. Specifically, packet inter-arrival data was collected and analyzed for a message passing interface (MPI) and an HTTP problem that simulated packet movement in an enterprise level LAN test-bed. The parallelization method that we employed broke the problem into N subparts based in the number of processors used. Inter-processor communication was required whenever a processor needed to interact with another one to help solve the application. That communication was relayed via the MPI protocol using TCP packets with intensities in .0005 second range. Experimental trials were run on various platforms: from 2 to 12 two processor units for MPI, and 8 and 16 units for HTTP. In the MPI experiment, although the CPU time continued to drop as additional units were added, the elapsed time only dropped to the 4 unit level and then increased thereafter. In the HTTP experiments the two switch/server model provided only a slightly better performance than the one switch/server model.

TU‐FF‐A1‐06: A Robust Scalable Parallel Processing System for Radiation Therapy

Purpose: To develop a robust Linux—based cluster for the parallel computation of problems of interest in a radiation therapy environment. This system should be robust, scalable and easy to manage. It should be constructed from commercially available low cost hardware and use only open source software tools to manage the system. Method and Materials: This cluster was constructed using a distributed memory model with the Message Passing Interface (MPI) protocol. The use of distributed memory requires a fast backbone to efficiently distribute programs and data to the various cluster nodes. This rapid data transfer was accomplished using a Gigabit Ethernet switch which allows a peak transfer rate of 100 Mbytes/sec. The cluster currently consists of 76 CPU's each with a minimum of 512 Mbytes of RAM. The individual nodes run an open source version (Centos 4.0) of the Redhat Enterprise 4.0 Linux operating system. The MPI protocol is implemented using the open source implementation, MPICH. Cluster node management is accomplished using the ROCKS 4.0 shareware toolset. The compilers and debuggers (C++ and FORTRAN for Linux) are Intel 9.0. Finally, the Integrated Development Environment (IDE) is the Eclipse open‐source project v3.0.1 with PHOTRAN extensions. Results: This cluster has recently been commissioned and several benchmark tests have been completed. Factors of 15X – 60X improvement in speed for parallelizable sections of various codes have been demonstrated. Conclusions: This system is robust enough to solve complex problems which were intractable with our previous computational tools. The demonstrated speed improvements will allow for the implementation of codes for problems such as: real‐time dose calculation, fast IMRT optimization, and the convolution of correlated CT datasets to account for patient motion.

Parallelization of Rotorcraft Aerodynamics Navier-Stokes Codes

The modie cation of unsteady three-dimensional Navier ‐Stokes codes for application on massively parallel and distributed computing environments is investigated. Previously, theEulermode of theNavier ‐Stokescode TURNS hasbeenparallelized. Fortheefe cientimplementation oftheNavier ‐StokesmodeofTURNS onparallelcomputing systems, several algorithmic changes should be made. The main modie cation is done on the implicit operator, lower‐upper symmetric Gauss ‐Seidel. Two new implicit operators are used because of convergence problems of traditional operators with high cell aspect ratio grids needed for viscous calculations. Results for Navier ‐Stokes cases are presented for various operators. The message passing interface protocol is used because of its portability to various parallel architectures. Nomenclature a = speed of sound e = total energy p = pressure Pr = Prandtl number Re = Reynolds number u;v;w = Cartesian velocities x; y;z = Cartesian coordinates ° = specie c heat ratio π = dynamic viscosity »;´;‡ = generalized coordinates Ω = density 9 = rotor azimuth angle

Parallelization of an ecological landscape model by functional decomposition

A functional scheme is described to parallelize computer simulations of grid-based ecological landscape models. The method is implemented using the Message Passing Interface protocol and is applied to the Everglades Landscape Vegetation Model. On a two-processor system, the speed-up is satisfactory and the overall performance of the program is competitive with traditional parallelization techniques such as geometrical decomposition. The method is discussed, timing information is provided for three different parallel machines, and some further developments are indicated.