Parallelized Research Articles

Sublinear time approximation schemes for makespan minimization on parallel machines

Sublinear time approximation schemes for makespan minimization on parallel machines

Extending TOUGH + HYDRATE with a parallel particle transport simulator: numerical investigation of sand production during gas production from hydrate deposits

Abstract A new parallel code for simulating particle transport in porous media is integrated with the TOUGH + HYDRATE simulator to investigate sand production associated with gas production from unconsolidated gas hydrate-bearing sediments (HBS). The parallel coupled simulator is named THMPT and uses the integral finite difference method to describe the Darcian and non-Darcian flow of fluids and heat transport, the finite element method to describe the associated geomechanical changes, and the discrete element method to track the trajectory of individual sand particles within the HBS. The THMPT simulator is written in Fortran, incorporates multiple optimized algorithms, and can comprehensively address the coupled flow, thermal, chemical, geomechanical, and particle transport processes that characterize the system behaviors during gas production from HBS. The simulator can capture all processes involved in sand particle transport in porous media, including sand detachment, collision, clogging (i.e., bridging), and migration. A benchmark case study of sand production in the course of depressurization-induced gas production from a representative HBS reveals various distinct microscopic particle migration mechanisms and the adverse impact of sand particle detachment, transport, and clogging. The numerical investigation also examines the effect of bottomhole pressure on mitigating sand production. The simulation results indicate that sand clogging near the wellbore significantly reduces permeability, decreasing gas production by at least 50%. Lastly, the efficiency of gravel packing in mitigating sand production is numerically evaluated, revealing that the structure of the porous media appears to profoundly influence the macroscopic motion behavior of sand particles and sand clogging characteristics.

Multi-wavelength diffractive optical neural network integrated with 2D photonic crystals for joint optical classification

Abstract Optical neural networks (ONNs) have demonstrated unique advantages in overcoming the limitations of traditional electronic computing through their inherent physical properties, including high parallelism, ultra-wide bandwidth, and low power consumption. As a crucial implementation of ONNs, on-chip diffractive optical neural network (DONN) offers an effective solution for achieving highly integrated and energy-efficient machine learning tasks. Notably, wavelength, as a fundamental degree of freedom in optical field manipulation, exhibits multidimensional multiplexing capabilities that can significantly enhance computational parallelism. However, existing DONNs predominantly operate under single-wavelength mechanisms, limiting the computational throughput. Here, we propose a multi-wavelength visual classification architecture termed PhC-DONN, which integrates two-dimensional photonic crystal (PhC) components with diffractive computing units. The architecture comprises three functional modules: (1) a PhC convolutional layer that enables multi-wavelength feature extraction; (2) a three-stage diffraction layer performing parallel modulation of optical fields; and (3) a PhC nonlinear activation layer implementing wavelength nonlinear computation. The results demonstrate that the PhC-DONN achieves classification accuracies of 99.09 % on the MNIST dataset, 66.41 % on the CIFAR-10 dataset, and 92.25 % on KTH human action recognition. By introducing a wavelength-parallel classification mechanism, the architecture accomplishes multi-channel inference during a single light propagation pass, resulting in a 32-fold enhancement in computational throughput compared to conventional DONNs while improving classification accuracy. This work not only establishes a novel optical classification paradigm for multi-wavelength optical neural network, but also provides a viable pathway towards constructing large-scale photonic intelligence parallel processors.

A new, efficient approach to speed up local search by estimating the solution quality: an application to stochastic, parallel machine scheduling

Local search has become a very powerful tool to solve hard optimization problems. One of the key parts here is that you must decide to accept or reject a new solution, for which we compare the objective values of the incumbent and the new solution. Since in general very many iterations are involved, it is important that this comparison can be done efficiently. In case of a stochastic problem, where we want to optimize the expected value, it is often impossible to do the comparison analytically. We can resolve this by simulating a solution, but to find an accurate estimate we need many simulations, which may take a lot of time. In this paper we present an alternative method to estimate the expected value for problems involving waiting relations. To apply this method we must iteratively compute the maximum of several stochastic variables. Thereto, we pretend that all stochastic variables are normally distributed, after which we iteratively estimate the expected value and standard deviation of the maximum of two variables; in the further analysis we pretend this maximum to be normally distributed again. We test the efficiency of this method on a specific scheduling problem with precedence relations. In our experiments we find that this approximation method in many cases produces better solutions than estimating the expected makespan using 1000 independent simulations per iteration, and it always dominates using 300 simulations per iteration, while using only a fraction of the time. Moreover, this method of estimating the distribution of the maximum seems to be widely applicable. For example, we can use it for all kinds of planning problems in which a person/task must wait until at least two other events have been completed, which is a typical situation in which a direct computation of the expected value is intractable.

A Parallel Implementation for Large-Scale TSR-based 3D Structural Comparisons of Protein and Amino Acid

Background: Proteins play a vital role in sustaining life, requiring the formation of specific 3D structures to manifest their essential biological functions. Structure comparison techniques are benefiting from the ever-expanding repositories of the Protein Data Bank. The development of computational tools for protein and amino acid 3D structural comparisons plays an important role in understanding protein functions. The Triangular Spatial Relationship (TSR)-based was developed for such purpose. Methods: A parallelization strategy and actual implementation on high-performance clusters using the distributed and shared memory programming model, along with the utilization of multi-core CPU and many-core GPU accelerators, were developed. 3D structures of proteins and amino acids are represented by an integer vector in the TSR-based method. This parallelization strategy is designed for the TSR-based method for large-scale 3D structural comparisons of proteins and amino acids in this study. It can also be adapted to other applications where a vector type of data structure is used. Results: Due to the nature of the vector representation of protein and amino acid structures using the TSR-based method, the comparison algorithm is well-suited for parallelization on large scale supercomputers. Performance studies on the representative datasets were conducted to demonstrate the efficiency of the parallelization strategy. It allows comparisons of large 3D protein or amino acid structure datasets to finish within a reasonable amount of time. Conclusion: The case studies, by taking advantage of this parallelization code, demonstrate that applying either mirror image or feature selection in the TSR-based algorithms improves the classifications of protein and amino acid 3D structures. The TSR keys have the advantage of performing structure-based BLAST searches. The parallelization code could be used as a reference for similar future studies.

A memetic algorithm based on feasible solutions with local search for fuzzy unrelated parallel machine scheduling with fuzzy cost constraint

A memetic algorithm based on feasible solutions with local search for fuzzy unrelated parallel machine scheduling with fuzzy cost constraint

Mathematical formulations for the scheduling of unrelated parallel machines considering machine availability and eligibility

Mathematical formulations for the scheduling of unrelated parallel machines considering machine availability and eligibility

Two new parameter proposals for variable neighborhood descent algorithm to minimize Cmax in machine scheduling problem

Scheduling can be defined as the assignment of jobs to machines that will be processed under certain constraints and measurements. Different scheduling problems arise when creating schedules and assigning jobs to machines. The problem of interest in this study is the Unrelated Parallel Machine Scheduling Problem with Sequence-Dependent Setup Times (UPMSPST), which is classified as NP-hard. The objective of the problem is to minimize the makespan (Cmax). In UPMSPST, machines have different processing times for jobs and there are machine-dependent setup times between jobs. Exact solution methods are not sufficient for solving the UPMSPST and many metaheuristics have been proposed by researchers to find approximate solutions. The aim of this paper is to propose two new parameters for solving the UPMSPST with the Variable Neighborhood Descent (VND) algorithm, which is a single-solution metaheuristic algorithm, in order to find better solutions. The proposed parameters are tested in four different scenarios and the results of the best parameter configuration are given. The results show that the proposed new parameters are effective in improving the existing solution.

Automated Sentence Alignment in Ukrainian-German Parallel Texts

Introduction. Sentence alignment in German parallel texts is a relevant task. It allows obtaining parallel data sets necessary for many computational linguistics tasks, such as parallel corpus construction and machine translation. The article describes the main tasks of sentence alignment, reviews existing methods and analyzes their ideas. Based on this analysis, a new method is proposed, it is based on the Bleualign approach, which uses machine translation systems and the BLEU metric to assess the similarity of sentences. However, it differs in the use of additional marker dictionaries for industry terms and conjunctions, including their synonyms. This article outlines the main tasks of sentence alignment, reviews existing methods, and discusses their ideas. Based on this analysis, a new method is proposed. This method is based on the Bleualign approach, which uses machine translation systems and BLEU metrics to evaluate sentence similarity, including the alignment of parts of complex sentences. However, it differs in the alignment process steps and introduces additional marker dictionaries for domain-specific words and conjunctions, including their synonyms. The purpose of the work is to develop a method and software for automated sentence alignment in Ukrainian-German parallel texts. Methods. The developed method is based on the Bleualign method and the BLEU metric. It is improved by the use of dictionaries of industry terms and conjunctions, and also provides a focus on one language pair — Ukrainian-German. The proposed method consists of 6 stages, allowing to align sentences in Ukrainian-German parallel texts. The proposed method is implemented in software using the Python programming language. Results. A new method for aligning sentences for Ukrainian-German parallel texts has been developed and its software implementation has been completed. The proposed method is based on statistical approaches and does not require significant computing resources. Unlike the Bleualign method, it uses dictionaries of industry terms and conjunctions for more accurate sentence alignment. Conclusions. Further research will include experiments and comparison of the alignment results obtained using the proposed method with the results of the Bleualign method.

Dominance Rules for Solving Scheduling on Uniform Parallel Machines with Multi Objective Function

The problem of uniform Parallel Machine Scheduling (UPMS) in manufacturing system is considered with utilizing the aims of minimizing total completion time, total tardiness and total earliness. Multi objective functions have been developed to select optimal solution in uniform parallel machine scheduling. In the present study, our focus is on the NP-hard problem for scheduling n jobs on a uniform parallel machine. We create the dominance properties are incorporated.

ARACHNE: Optimizing Distributed Parallel Applications with Reduced Inter-Process Communication

In high-performance computing (HPC), parallelization is essential for improving computational efficiency as data and computation scales exceed single-node capacity. Existing methods, such as the polyhedral model used in Pluto -Distmem, focus on loop and array optimizations within shared memory but struggle with high communication overheads and inflexibility in distributed environments. These methods often fail to effectively partition computation and manage data across nodes, leading to suboptimal performance. This paper presents Arachne , an innovative system designed to address these shortcomings by generating distributed parallel code with minimized communication overhead. The system introduces a dynamic programming algorithm to optimally distribute computational tasks across multiple processes, ensuring minimal communication costs. It also incorporates user-friendly compiler directives, allowing programmers to influence code generation easily and accommodate a broader range of parallelization scenarios without needing in-depth knowledge of parallel architectures. Arachne significantly reduces the learning curve and need for extensive code modifications, making parallel programming more accessible and efficient. Evaluation of various HPC benchmarks demonstrates that Arachne outperforms existing methods by reducing communication overhead, lowering memory requirements, and supporting more complex parallel logic, thus enhancing the overall scalability and efficiency of HPC applications.

Modeling of parallel program synchronization primitives

This article is devoted to the problem of verifying parallel programs that may contain special types of errors associated with the synchronization of parallel executed threads and access to shared memory. Such errors include deadlocks and data races. There is a division of parallel program verification methods into static and dynamic. The second ones require running the code and allow to check only the current implementation of the program for races, which, if there are a large number of branches, can lead to missing races. Among static methods, analytical methods (for example, based on deductive analysis) and model checking methods are most widely used. However, they are difficult to implement, and model checking still require a significant amount of manual work from the programmer to build such a model. In this regard, it is necessary to use models that can be built automatically. Previously, the authors developed a model based on an extension of Petri nets, which allows automatic creation based on sequential code and its conversion into parallel code. Automatic creation of a model of a parallel program introduces new, previously unused requirements related to the interaction of parallel threads. Thus, this article discusses the features of modeling using extended Petri nets with semantic relations of the main synchronization primitives implemented in most languages and parallel programming technologies for shared memory systems. In the future, these models will be used to search for data races and deadlocks in parallel programs.

Optimising batch scheduling on non-identical parallel machines: lower bounds, MIP, branch-and-price, and heuristics

This paper addresses a scheduling problem involving a set of diverse jobs processed in batches. Each job has its own processing time and size. Notably, the processing time of any batch is determined by the longest processing time among the jobs it contains. These batches are scheduled on parallel machines with non-identical capacities, ensuring that the total size of the jobs in any batch does not exceed the capacity of the assigned machine. The objective is to minimise the maximum makespan across all machines, where the makespan is defined as the sum of the processing times of all batches on a machine. To tackle this problem, the paper proposes lower bounds, a mixed-integer programming formulation, and a branch-and-price algorithm. Additionally, five rule-based heuristics and a beam search algorithm are introduced. An extensive computational study demonstrates that the proposed solution methods outperform existing approaches in the literature in terms of both solution quality and computational efficiency. Furthermore, the proposed methods achieve best-known solutions for the first time in the literature, demonstrating their significant improvement over existing approaches.

Greedy Minimum-Energy Scheduling

We consider the problem of energy-efficient scheduling across multiple processors with a power-down mechanism. In this setting a set of n jobs with individual release times, deadlines, and processing volumes must be scheduled across m parallel processors while minimizing the consumed energy. When idle, each processor can be turned off to save energy, while turning it on requires a fixed amount of energy. For the special case of a single processor, the greedy Left-to-Right algorithm [1] guarantees an approximation factor of 2. We generalize this simple greedy policy to the case of processors running in parallel and show that the energy costs are still bounded by , where is the energy consumed by an optimal solution and is the total processing volume. Our algorithm has a running time of , where d is the difference between the last deadline and the earliest release time, and f is the running time of a maximum flow calculation in a network of nodes.

Dynamically Fusing Python HPC Kernels

Recent trends in high-performance computing show an increase in the adoption of performance portable frameworks such as Kokkos and interpreted languages such as Python. PyKokkos follows these trends and enables programmers to write performance-portable kernels in Python which greatly increases productivity. One issue that programmers still face is how to organize parallel code, as splitting code into separate kernels simplifies testing and debugging but may result in suboptimal performance. To enable programmers to organize kernels in any way they prefer while ensuring good performance, we present PyFuser, a program analysis framework for automatic fusion of performance portable PyKokkos kernels. PyFuser dynamically traces kernel calls and lazily fuses them once the result is requested by the application. PyFuser generates fused kernels that execute faster due to better reuse of data, improved compiler optimizations, and reduced kernel launch overhead, while not requiring any changes to existing PyKokkos code. We also introduce automated code transformations that further optimize the fused kernels generated by PyFuser. Our experiments show that on average PyFuser achieves speedups compared to unfused kernels of 3.8x on NVIDIA and AMD GPUs, as well as Intel and AMD CPUs.

A Heuristic Mutation Based Genetic Algorithm for Fast Parallel Scheduling of Steel Cold Rolling

A well-designed production schedule for cold rolling can enhance steel enterprises' operational efficiency and profitability. Nevertheless, the intricate constraints and numerous steps involved in cold rolling pose challenges to devising a rational scheduling plan. Therefore, considering the practical production constraints, this paper investigates a cold rolling scheduling problem for processing jobs with specific due dates and batch attributions on parallel heterogeneous machines with continuous production requirements. Firstly, the scheduling problem is formulated as a mixed integer linear program (MILP) model with an economic objective. Then, a modified genetic algorithm (GA) is proposed to search for the optimal solution to the MILP problem. Specifically, this method includes a heuristic initialization mechanism to generate feasible initial solutions, three heuristic mutation operators to generate promising candidate solutions, and a parallel computing mechanism to accelerate the evaluation process of the GA. The simulation results demonstrate that the proposed method can be effectively implemented to generate optimized scheduling schemes in the cold rolling process.

Reverse Engineering of Music Mixing Graphs With Differentiable Processors and Iterative Pruning

Reverse engineering of music mixes aims to uncover how dry source signals are processed and combined to produce a final mix. In this paper, prior works are extended to reflect the compositional nature of mixing and search for a graph of audio processors. First, a mixing console is constructed, applying all available processors to every track and subgroup. With differentiable processor implementations, their parameters are optimized with gradient descent. Next, the process of removing negligible processors and fine-tuning the remaining ones is repeated. This way, the quality of the full mixing console can be preserved while removing approximately two-thirds of the processors. The proposed method can be used not only to analyze individual music mixes but also to collect large-scale graph data for downstream tasks such as automatic mixing. Especially for the latter purpose, efficient implementation of the search is crucial. To this end, an efficient batch-processing method that computes multiple processors in parallel is presented. Also, the “dry/wet” parameter of each processor is exploited to accelerate the search. Extensive quantitative and qualitative analyses are conducted to evaluate the proposed method’s performance, behavior, and computational cost.

Configuration synthesis of variable-mode parallel machining mechanisms based on FIS theory

Configuration synthesis of variable-mode parallel machining mechanisms based on FIS theory

Attribution of the 2023 extreme spring hot drought event in Southwest China: meteorological and agricultural perspectives

Abstract During March to May 2023, Southwestern China experienced a prolonged hot and dry spring. This study investigates the role of anthropogenic climate change (ACC) in this extreme event and its impact on crop yields, using a 525-member ensemble from the HadGEM3 atmospheric model and the WOFOST agricultural model, respectively. Observational results indicate that the area-averaged surface air temperature anomaly (TAS) was 1.4 °C higher than usual, and the percentage of anomalous precipitation was nearly 33% below normal, making both metrics the most extreme since 1960. The hot event was predominantly driven by ACC, contributing about 60% to the TAS strength and increasing its likelihood 7 fold. In contrast, internal climate variability appears to have been the primary trigger of the severe drought, although ACC significantly increased its likelihood by 9 fold. When these two extreme conditions are considered together as a concurrent hot-drought event, ACC increased its probability by about 3 fold. An attribution analysis of crop yields in the area was also conducted, revealing that ACC significantly shifts the probability distribution westward and reduces yields for both the winter wheat and rapeseed. Specifically, based on the simulated 2023 crop yield results driven by ERA5 as a proxy for observations, the likelihood of achieving the yield for winter wheat would have decreased by a factors of about 2 under ACC, while that for winter rapeseed is close to 1, despite its better simulation performance in the historical context. This underscores the challenge of accurately modeling and projecting crop yields under changing climate conditions. Additional analysis emphasizes the role of land–atmosphere interactions in amplifying the hot-drought event, as early-season soil moisture deficits reduced evapotranspiration and increased sensible heat flux, thereby intensifying local temperature extremes. Overall, global warming caused by anthropogenic activities has significantly increased the frequency of extreme hot drought events in Southwestern China, posing a severe threat to agricultural productivity.

A scalable distributed pipeline for reference-free variants calling

BackgroundPrecision medicine pipelines typically begin with variant calling to identify disease-related mutations for optimal treatment selection. Reference-free approaches assess variations in the genetic profiles of distinct individuals through the utilization of a De Bruijn graph. However, the timely analysis of large-scale sequencing data may be beyond the capabilities of single workstations, requiring alternative computational approaches.ResultsWe introduce the first-known distributed pipeline for detecting isolated SNPs (Single Nucleotide Polymorphisms), by leveraging the computational resources of multiple machines in parallel. Our pipeline efficiently analyzes large datasets thanks to the usage of a distributed De Bruijn graph representation. Furthermore, we introduce a cluster-driven algorithm to partition the De Bruijn graph across multiple independent machines according to the inner structure of the sequences under analysis, thus further improving the scalability of our pipeline.ConclusionsThe results of our experiments, conducted on real-world datasets, show the good performance of our pipeline in terms of efficiency, output quality and scalability. Moreover, the reported results also confirm that the adoption of a specialized partitioning algorithm for the distributed representation of the De Bruijn graph leads to a relevant performance speed-up compared to using standard partitioning techniques.