Supercomputer Center Research Articles

Modeling spatial patterns in a moving crowd of people using data-driven approach—A concept of Interplay Floor Field

On the basis of data-driven methodology, especially velocity correlations of pedestrians moving in a crowd, we have proposed a new model of pedestrian dynamics with an easy-adjustable space discretization. The model is based on Cellular Automata (CA) with an adaptive lattice and takes into account proxemics patterns among pedestrians. The proposed model uses agents located on the CA lattice, constructed using the concept of Floor Field (FF), namely, a set of gradient potential fields influencing the movement of agents. In the model, we have proposed three kinds of such fields: Static FF—responsible for navigation to agents’ Points of Interest (POIs), Wall FF—a repulsive influence with obstacles, and Interplay FF—to model agents’ volume and proxemics effects. The third field, which models mutual relations between pedestrians taking into account the rules of proxemics is an added value in the area of crowd modeling and simulation.During the creation of the proposed model, we have based on a set of experimental data provided by Jülich Supercomputing Centre and our previous analysis of spatial patterns and velocity correlations. Thus, the proposed model includes, among other things the most probable positions of the nearest neighbors in a crowd—including the shape of “the forbidden zone” around a pedestrian and the finding of rules explaining observed mutual spatial relations between neighbors in a crowd. In the described model the cell size is an adjustable parameter, with available values from 1 cm to 50 cm. It is possible with the usage of embodied agents which always occupy only one cell of Cellular Automata, actual pedestrian volume is modeled with the Interplay FF. An interesting property of the model is that changing the cell size does not involve changing the rules of the model, e.g. its transition function.

A Multilevel Extension of the GDSW Overlapping Schwarz Preconditioner in Two Dimensions

Abstract Multilevel extensions of overlapping Schwarz domain decomposition preconditioners of Generalized Dryja–Smith–Widlund (GDSW) type are considered in this paper. The original GDSW preconditioner is a two-level overlapping Schwarz domain decomposition preconditioner, which can be constructed algebraically from the fully assembled stiffness matrix. The FROSch software, which belongs to the ShyLU package of the Trilinos software library, provides parallel implementations of different variants of GDSW preconditioners. The coarse problem can limit the parallel scalability of two-level GDSW preconditioners. As a remedy, in the past, three-level GDSW approaches have been proposed, which can significantly extend the range of scalability. Here, a multilevel extension of the GDSW preconditioner is introduced and analyzed. Finally, parallel results for the implementation in FROSch for up to 40 000 cores of the SuperMUC-NG supercomputer at Leibniz Supercomputing Centre (LRZ) and to 48 000 cores of the JUWELS supercomputer at Jülich Supercomputing Centre (JSC) are presented.

Deciphering the dynamics of distorted turbulent flows: Lagrangian particle tracking and chaos prediction through transformer-based deep learning models

Turbulent flow is a complex and vital phenomenon in fluid dynamics, as it is the most common type of flow in both natural and artificial systems. Traditional methods of studying turbulent flow, such as computational fluid dynamics and experiments, have limitations such as high computational costs, experiment costs, and restricted problem scales and sizes. Recently, artificial intelligence has provided a new avenue for examining turbulent flow, which can help improve our understanding of its flow features and physics in various applications. Strained turbulent flow, which occurs in the presence of gravity in situations such as combustion chambers and shear flow, is one such case. This study proposes a novel data-driven transformer model to predict the velocity field of turbulent flow, building on the success of this deep sequential learning technique in areas such as language translation and music. The present study applied this model to experimental work by Hassanian et al., who studied distorted turbulent flow with a specific range of Taylor microscale Reynolds numbers 100<Reλ<120. The flow underwent a vertical mean strain rate of 8 s−1 in the presence of gravity. The Lagrangian particle tracking technique recorded every tracer particle's velocity field and displacement. Using this dataset, the transformer model was trained with different ratios of data and used to predict the velocity of the following period. The model's predictions significantly matched the experimental test data, with a mean absolute error of 0.002–0.003 and an R2 score of 0.98. Furthermore, the model demonstrated its ability to maintain high predictive performance with less training data, showcasing its potential to predict future turbulent flow velocity with fewer computational resources. To assess the model, it has been compared to the long short-term memory and gated recurrent units model. High-performance computing machines, such as JUWELS-DevelBOOSTER at the Juelich Supercomputing Center, were used to train and run the model for inference.

Информационно-вычислительные ресурсы ИРНОК: инфраструктура, данные, приложения

В статье представлены этапы развития информационно-вычислительной и телекоммуникационной инфраструктуры Иркутского научно-образовательного комплекса в период с 2008 по 2022 г. Показано развитие инфраструктуры: сети передачи данных, систем хранения данных, виртуализации, вычислительных ресурсов, в том числе оборудования Иркутского суперкомпьютерного центра. Abstract The Integrated information and computing network of the Irkutsk scientific and educational complex (IICN IREC) is the basic technological platform for most interdisciplinary research conducted by scientific organizations of the Irkutsk region. The article presents the stages of development of the information, computing and telecommunications infrastructure of the IICN IREC in the period from 2008 to 2022. The main directions of modernization of IREC were: the development of a communication network, the deployment of a virtualization platform, data storage system and GPU cluster. The Irkutsk supercomputer center (ISCC) of the SB RAS, which interacts with IREC, has also been developed. The most significant stage in the development of the ISCC was the implementation of the project to create a new generation computing cluster, named after the first organizing director of the institute, Academician V.M. Matrosov. The effectiveness of the created information and computing infrastructure is shown on the example of the implementation of two major integration projects. These are the projects “Fundamental research and breakthrough technologies as the basis for the advanced development of the Baikal region and its interregional relations” (2016–2019) and “Fundamental foundations, methods and technologies for digital monitoring and forecasting of the environmental situation in the Baikal natural territory (BNT)” (2020–2022). More than a dozen institutions have taken part in these projects. To support interdisciplinary research, tools that ensure effective interaction of researchers with the IREC information and computing infrastructure have been developed at the ISDCT SB RAS. Among them are Geoportal, Digital Platform for Monitoring of BNT and Orlando Tools. As part of the projects, a large array of various formats data on the state of the region was collected and computationally complex problems were solved. The integration of heterogeneous scientific data and the possibility of processing them with high-performance computing tools open up new opportunities for obtaining new breakthrough results.

Large scale performance analysis of distributed deep learning frameworks for convolutional neural networks

Continuously increasing data volumes from multiple sources, such as simulation and experimental measurements, demand efficient algorithms for an analysis within a realistic timeframe. Deep learning models have proven to be capable of understanding and analyzing large quantities of data with high accuracy. However, training them on massive datasets remains a challenge and requires distributed learning exploiting High-Performance Computing systems. This study presents a comprehensive analysis and comparison of three well-established distributed deep learning frameworks—Horovod, DeepSpeed, and Distributed Data Parallel by PyTorch—with a focus on their runtime performance and scalability. Additionally, the performance of two data loaders, the native PyTorch data loader and the DALI data loader by NVIDIA, is investigated. To evaluate these frameworks and data loaders, three standard ResNet architectures with 50, 101, and 152 layers are tested using the ImageNet dataset. The impact of different learning rate schedulers on validation accuracy is also assessed. The novel contribution lies in the detailed analysis and comparison of these frameworks and data loaders on the state-of-the-art Jülich Wizard for European Leadership Science (JUWELS) Booster system at the Jülich Supercomputing Centre, using up to 1024 A100 NVIDIA GPUs in parallel. Findings show that the DALI data loader significantly reduces the overall runtime of ResNet50 from more than 12 h on 4 GPUs to less than 200 s on 1024 GPUs. The outcomes of this work highlight the potential impact of distributed deep learning using efficient tools on accelerating scientific discoveries and data-driven applications.Graphical

Pinpointing atrial fibrillation drivers by informed sequential endocardial mapping in the presence and absence of fibrosis: an in-silico study

Abstract Funding Acknowledgements Type of funding sources: Public grant(s) – EU funding. Main funding source(s): This work is part of Personalize AF. This project received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 860974. This work was also supported by the Swiss National Supercomputing Centre (CSCS), project s1074. Background We recently developed a novel technique to localize spatially stable atrial fibrillation (AF) drivers using the combined information from sequential endocardial mapping recordings. The robustness of this technique to localize drivers in the presence of complex substrate with a high degree of fibrosis and with unstable and meandering drivers is not known. Purpose To evaluate the effect of spatial instability of drivers and high AF substrate complexity caused by atrial fibrosis on the performance of the recently developed technique for AF driver localisation using sequential endocardial mapping. Methods We used a previously validated volumetric 3D atrial model to simulate AF initiated by incremental pacing. Two sets of simulation were generated, with no structural changes (n=7), and with severe remodelling (n=11), implemented as patches of endomysial fibrosis covering 70% of the atria. Simulated endocardial unipolar electrograms were generated sequentially in a 4x4 electrode grid (3 mm spacing, 2.5 s) starting from 20 uniformly spaced positions in the atria. The subsequent catheter positions were determined by analysing the propagation direction in the grid. If the measured activity was repetitive, the dominant direction of propagation was determined, and the catheter was repositioned 10 mm upstream within the same atrium. Otherwise, the catheter was moved 5 mm in a random direction. This procedure continued until either 1) a driver (re-entry or radial spread of activations) was detected directly at the recorded region by evaluating the corresponding propagation patterns, or 2) the previous catheter positions defined an enclosed region in the atrium, encircling the AF driver. The performance of the algorithm was compared between the two remodelling stages with respect to number of steps required to locate a driver, and to the distance to the closest driver. Results In the simulations with fibrosis more drivers were present than in the simulations without fibrosis (4 [2, 6] vs. 3 [2; 4] simultaneous reentries, p < 0.001). Drivers were spatially unstable and meandered substantially through the atrium. The performance of the algorithm to localize the drivers was similar in both groups. The tracking procedures located a driver after 6 [IQR: 4; 9] vs. 5 [3; 8] steps (p<0.01). The algorithm localized the driver correctly in 72.8% vs. 78.2% (p = 0.30) of the cases and the final distances to a driver were 7.87 [5.88; 11.04] mm and 7.78 [5.55, 11.35] mm (p=0.95), respectively. More drivers were detected by encircling them with sequential catheter positions in the cases without remodelling (62.6% vs. 40.9%, p<0.001). Conclusion Guiding mapping catheters by functional aspects of AF and combining sequential recordings may increase the efficiency and accuracy of mapping procedures. The developed algorithm successfully localizes AF drivers even in complex substrates for AF.

Detection of the most dominant drivers of atrial fibrillation based on repetitive activity in sequential electro-anatomical mapping

Abstract Funding Acknowledgements Type of funding sources: Public grant(s) – EU funding. Main funding source(s): This work is part of Personalize AF. This project received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 860974, This work was also supported by the Swiss National Supercomputing Centre (CSCS), project s1074. Introduction Detecting and targeting the most dominant drivers of atrial fibrillation (AF) might improve the outcome of AF ablation procedures. We hypothesize that dominant drivers entrain their vicinity in a temporally and spatially stable manner resulting in local as well as adjacent repetitive activation patterns. Purpose We present a novel approach to detect the most dominant drivers of AF based on driver repetitiveness and the degree to which they entrain their vicinity. Methods We retrospectively analyzed a dataset of high-density bi-atrial sequential mapping in ablation-naive persistent AF patients (n=13, 30s recording per site). Candidate drivers (focal activity or re-entry) were detected based on local activation time analysis of unipolar electrograms. Focal activity was defined as radial spread of activation, re-entry as a circular activation sequence on at least 4 splines of the mapping catheter that covered >50% of the AF cycle length. As a measure of driver stability, we determined the driver's local repetitiveness (in % of recording). As a measure of driver-associated entrainment, we determined the directionally coherent repetitiveness of adjacent recordings within 30 mm (Figure 1). Following our hypothesis, most dominant drivers can be recognized by their high local and directional coherent adjacent repetitiveness. Results A total of 459 recordings were analyzed (35±5 per patient). We detected 130 focal activities (10±3 per patient) and 42 re-entries (4±2 per patient) in total. The spatial distribution of all drivers is shown in Figure 2 (blue bars). Focal activities were more repetitive than re-entries (Figure 2, median [IQR] 3.2% [1.9%, 5.6%] vs. 1.8% [1.8%, 2.2%], p<0.001 Mann-Whitney U-test). When applying a 90th percentile threshold (8.2%) to both the local and directional coherent adjacent repetitiveness to detect the most dominant drivers, we were able to detect 9 dominant drivers (all focal activities, 7 patients). Most dominant drivers were located in the upper right atrium or left pulmonary vein region (Figure 2). Interestingly, only 9 of the 130 initially detected drivers (7%) showed an activation pattern in line with a repetitive entrainment of their vicinity. Furthermore, in 6 patients (46%), no dominant drivers were detected. This might indicate that not all patients have temporally and spatially stable drivers. Conclusion In this abstract we present a novel methodology to detect the most dominant drivers of AF. Using this method, we found that some (but not all) patients have temporally and spatially stable drivers that are able to entrain their vicinity.

Liquid Super-Spreading Boosted High-Performance Jet-Flow Boiling for Enhancement of Phase-Change Cooling.

Enhanced boiling heat transfer via surface engineering is a topic of general interest for its great demand in industrial fields. However, as a dynamic interfacial phenomenon, deep understanding of its process and mechanism, including liquid re-wetting and vapor departure, is still challenging. Herein, we design a micro-/nanostuctured Cu surface containing periodic micro-groove/pyramid array with rich nano-wrinkles, where super-spreading (<134.1ms) of organic cooling agents highly boosts liquid re-wetting process, causing a discontinuous solid-liquid-vapor three-phase contact line and ultra-low under-liquid bubble adhesion force (∼1.3 μN). Therefore, a characteristic, ultra-fast jet-flow boiling (bubbles rapidly ejected in multiple strips) is obtained on this surface, giving a priority to nucleation (superheat ∼1.5°C) and simultaneously enhancing the critical heat flux and heat-transfer coefficient by up to 80% and 608%, respectively, compared with a flat surface. In situ observation and analysis of the nucleation, growth, and departure of micro-sized jet-flow bubbles reflects that micro-grooves/pyramids with nano-wrinkles promote the latent heat exchange process by super-spreading induced ultra-fast liquid re-wetting and constant vapor film coalescing. Based on the designed structures, high-performance phase-change cooling for CPU heat management in supercomputer centers is accomplished with an ultra-low power usage effectiveness (PUE<1.04). This article is protected by copyright. All rights reserved.

From the chair

This first issue of 2023 picks up where we left off with follow-ups from our flagship conference, SC22. There, SIGHPC Awards Chair, Bernd Mohr (Jülich Supercomputing Center), and I met Didem Unat (Koç University), the winner of the 2021 SIGHPC Emerging Woman Leader in Technical Computing (EWL/TC) Award. Dr. Unat was unable to travel to receive the award in SC21 due to you-know-what, so we remedied that at SC22. Speaking of, the call for the 2023 EWL/TC Award is now open with nominations due by June 30. Please take this opportunity to nominate a colleague!

PASC23 Scales to New Heights, June 26--28

The Platform for Advanced Scientific Computing (PASC) Conference, co-sponsored by SIGHPC and the Swiss National Supercomputing Centre (CSCS), takes place from June 26 to 28, 2023 at the Congress Center Davos, located in Davos, Switzerland. This year's theme is Computing Across Scales, Domains, and Communities.

Cluster Partition Operation Study of Air-Cooled Fan Groups in a Natural Wind Disturbance

This study discusses the influence of natural wind on the air flow of air-cooled condensers (ACCs) and then proposes a partition speed-regulation strategy for a fan group with enhanced generalized capability, which is of great practical significance for optimizing energy-saving operations. The stochastic time-varying features of natural wind are characterized by sine–Gaussian, Weibull, and composed winds. In a natural wind disturbance, using the Sugon Supercomputing Center, the transient numerical simulation of the dynamic evolution of the ACC flow field was found: the dynamic system of air flow is a typical time-varying nonlinear process. Cluster analysis was used to extract the nonlinear features of air flow, divide the fan group into four subregions with generalization capability, and implement a partitioned speed operation. It was found that giving priority to increasing the fan speed in the headwind partition can suppress the natural wind disturbance and improve the overall air flow, thus reducing the fan speed in the leeward partition, which reduces the overall air flow loss. The dynamic characteristics of the fan group obtained from the simulation and the proposed fan partition method can guide the optimized energy-saving operation of ACCs.

The Electric Dial-a-Ride Problem on a Fixed Circuit

Shared mobility services involving electric autonomous shuttles have increasingly been implemented in recent years. Because of various restrictions, these services are currently offered on fixed circuits and operated with fixed schedules. This study introduces a service variant with flexible stopping patterns and schedules. Specifically, in the electric dial-a-ride problem on a fixed circuit (eDARP-FC), a fleet of capacitated electric shuttles operates on a given circuit consisting of a recharging depot and a sequence of stations where passengers can be picked up and dropped off. The shuttles may perform multiple laps, between which they may need to recharge. The goal of the problem is to determine the vehicles’ stopping sequences and schedules, including recharging plans, so as to minimize a weighted sum of the total passenger excess time and the total number of laps. The eDARP-FC is formulated as a nonstandard lap-based mixed integer linear programming and is shown to be NP-Hard. Efficient polynomial time algorithms are devised for two special scheduling subproblems. These algorithms and several heuristics are then applied as subroutines within a large neighborhood search metaheuristic. Experiments on instances derived from a real-life system demonstrate that the flexible service results in a 32%–75% decrease in the excess time at the same operational costs. Funding: This work was supported by the Fonds Wetenschappelijk Onderzoek [Project Data-Driven Logistics: Grant S007318N; Project Optimizing the Design of a Hybrid Urban Mobility System: Grant G020222N; and Grant OR4Logistics]. Y. Molenbruch is partially funded by the Fonds Wetenschappelijk Onderzoek [Grant 1202719N]. The computational resources and services used in this work were provided by the Flemish Supercomputer Center funded by the Fonds Wetenschappelijk Onderzoek and the Flemish Government. Supplemental Material: The electronic companion is available at https://doi.org/10.1287/trsc.2023.1208 .

Versatile software-defined HPC and cloud clusters on Alps supercomputer for diverse workflows

Supercomputers have been driving innovations for performance and scaling benefiting several scientific applications for the past few decades. Yet their ecosystems remain virtually unchanged when it comes to integrating distributed data-driven workflows, primarily due to rather rigid access methods and restricted configuration management options. X-as-a-Service model of cloud has introduced, among other features, a developer-centric DevOps approach empowering developers of infrastructure, platform to software artefacts, which, unfortunately contemporary supercomputers still lack. We introduce vClusters (versatile software-defined clusters), which is based on Infrastructure-as-code (IaC) technology. vClusters approach is a unique fusion of HPC and cloud technologies resulting in a software-defined, multi-tenant cluster on a supercomputing ecosystem, that, together with software-defined storage, enable DevOps for complex, data-driven workflows like grid middleware, alongside a classic HPC platform. IaC has been a commonplace in cloud computing, however, it lacked adoption within multi-Petascale ecosystems due to concerns related to performance and interoperability with classic HPC data centres’ ecosystems. We present an overview of the Swiss National Supercomputing Centre’s flagship Alps ecosystem as an implementation target for vClusters for HPC and data-driven workflows. Alps is based on the Cray-HPE Shasta EX supercomputing platform that includes an IaC compliant, microservices architecture (MSA) management system, which we leverage for demonstrating vClusters usage for our diverse operational workflows. We provide implementation details of two operational vClusters platforms: a classic HPC platform that is used predominantly by hundreds of users running thousands of large-scale numerical simulations batch jobs; and a widely used, data-intensive, Grid computing middleware platform used for CERN Worldwide LHC Computing Grid (WLCG) operations. The resulting solution showcases reuse and reduction of common configuration recipes across vCluster implementations, minimising operational change management overheads while introducing flexibility for managing artefacts for DevOps required by diverse workflows.

Global Virtual Data Space for Unified Data Access Across Supercomputing Centers

In the wide-area high-performance computing environment, heterogeneous storage resources are geographically distributed in different supercomputing centers, which leads to the barriers between applications and data. This paper proposes a global virtual data space, named GVDS, to meet the needs of unified data access across supercomputing centers. GVDS integrates the parallel/distributed file systems of supercomputing centers to present a virtual space with tremendous storage capability for users. GVDS organizes users into groups for easy management, which allows users to share, collaborate, and perform computations on the stored data. For failure tolerance, global metadata is replicated and distributed on multiple supercomputing centers, redundant I/O service components are deployed in each supercomputing center. GVDS uses adaptive prefetching, caching, and request merging to improve access performance. Experimental results running on real-world supercomputing centers show that, GVDS can deliver excellent I/O performance running micro-benchmark, real-world traces and applications.

Computational Optimization of a Loosely-Coupled Strategy for Scale-Resolving CHT CFD Simulation of Gas Turbine Combustors

The accurate prediction of heat fluxes and, thus, metal wall temperatures of gas turbine (GT) combustor liners is a complicated and numerically expensive task. Computational Fluid Dynamics (CFD) support for the design of cooling systems is essential to ensure safe and proper operation of the entire gas turbine engine. Indeed, it is well known how complicated, and, at the same time, expensive it is to carry out experimental campaigns inside combustors operating under working conditions, and, therefore, pressurized and having high temperatures. The correct prediction of thermal fluxes in a CFD simulation depends on the proper modeling of all the involved phenomena and their interactions with each other. For this reason, Conjugate Heat Transfer (CHT) simulations are mandatory in gas turbine cooling system applications. Multiphysics and multiscale simulations, based on loosely-coupled approaches, have emerged as extremely effective numerical tools, providing enormous computational time savings, as compared with standard CHT simulations. The fundamental advantage of such approaches is based on the fact that each heat transfer mechanism is solved with the most suitable numerical setup, which leads to the use of spatial and temporal resolutions following the characteristic time scales of each phenomenon to be solved. For industrial applications, where the availability of numerical resources is limited and, at the same time, the timelines with which to obtain results are rather tight, having robust and easy-to-use loosely-coupled solutions available for the design of combustion chamber cooling systems would be extremely valuable. In this context, the objective of this work was to perform an initial optimization step for the multiphysics and multiscale tool, U-THERM3D, developed at the University of Florence to revise the coupling strategy workflow with a view to making the numerical tool faster and easier to use. The revised methodology was applied to the RSM gas turbine combustor model test case developed with cooperation between the Universities of Darmstadt, Heidelberg, Karlsruhe, and the DLR. In particular, all experimental tests were conducted by the Institute of Reactive Flows and Diagnostics (Reaktive Strömungen und Messtechnik) of the Department of Mechanical Engineering at TU Darmstadt, from which the gas turbine combustor model takes its name. The newly obtained results were compared and analyzed, both qualitatively and in terms of computational time savings, with those previously achieved with the current version of the U-THERM3D tool already studied by the authors and available in the literature. Moreover, an analysis of computing times was carried out relative to the super-computing center used for the different adopted methodologies.

On Problems in OpenBLAS Library Usage in Productized Code on RISC-V

The boundary element method usage for the numerical simulation in continuum mechanics problems leads to the need to solve a system of linear algebraic equations with a dense matrix. The de facto standards for the interface of functions over dense matrices and vectors software implementations are BLAS/LAPACK. Among the optimized open-source BLAS/LAPACK implementations, only the OpenBLAS library includes optimizations for the widest range of hardware platforms. This library is optimized for Intel, AMD, ARM and RISC-V architectures. The open RISC-V architecture ecosystem is currently actively developing. European supercomputing centers have opened RISC-V competence centers as part of the government's EuroHPC grant support, since solutions based on the ARM architecture are not recognized as part of the European initiative to develop its own technological independence. Currently, companies included in the international RISC-V consortium are developing not only high-performance RISC-V processors, but also AI accelerators, as well as video cards based on RISC-V architecture. OpenBLAS is actively supported and optimized for emerging RISC-V hardware and extensions. However, libraries used in product code are traditionally subject to strict requirements for stability and reliability in order to minimize possible errors and failures in the product. As it turned out, from this point of view, OpenBLAS has a number of problems that we had to solve in order to productize this library. In this article the OpenBLAS test system is described, the problems of testing the LAPACK functionality of the library and ways to solve them are discussed. In addition, the test coverage of the BLAS functionality is analyzed and the results achieved in increasing it are presented. It is planned to contribute the described changes to the OpenBLAS project.

Long-term demographic forecasting

The results of the latest demographic forecasts from the world’s leading specialized centers (United Nations Population Division, the Wittgenstein Center for Demography and Global Human Capital, the Institute for Health Metrics and Evaluation) are considered, demonstrating a certain bias in favor of individual countries and their calculation methods. The second part of this article provides a description of a digital twin of the planet’s demographic system constructed by a Chinese−Russian team and implemented in China’s national supercomputer center. In addition, the results of some calculations carried out using this tool are described.

Design and Implementation of Burst Buffer Over-Subscription Scheme for HPC Storage Systems

Burst Buffer is widely used in supercomputer centers to bridge the performance gap between computational power and the high-performance I/O systems. The primary role of Burst Buffer is to temporarily absorb the bursty I/O and reduce the heavy access on Parallel File System (PFS). However, the job resource manager on High-Performance Computer (HPC) systems prefers to use a dedicated Burst Buffer allocation approach, which eventually leads to the severely underutilized Burst Buffer resource. To improve the efficiency of using the expensive Burst Buffer resource, we analyze the I/O patterns on Burst Buffer in depth. We propose Burst Buffer over-subscription allocation method, which improves Burst Buffer utilization by allowing each job to access Burst Buffer only during its I/O phases so that the jobs can overlap each other. Furthermore, we develop a new I/O congestion-aware scheduler and a transparent data management system between Burst Buffer and PFS. Our approach also reduces the memory overhead and improves the data persistence of the data management system by adapting the persistent memory. With the proposed approach, not only the Burst Buffer utilization can be improved, but also HPC applications can achieve high I/O performance by exploiting the powerful Burst Buffer hardware capabilities. Experimental results show that BBOS can improve Burst Buffer utilization by up to 120% while more stable and higher checkpoint performance is guaranteed even under high I/O loads compared to other state-of-the-art schedulers. Besides, our approach can improve the hit ratio of restart requests by up to 96.4% and provides up to 210% higher restart throughput on Burst Buffer.

Calculation of the high-energy neutron flux for anticipating errors and recovery techniques in exascale supercomputer centres

The age of exascale computing has arrived and the risks associated with neutron and other atmospheric radiation are becoming more critical as the computing power increases, hence, the expected Mean Time Between Failures will be reduced because of this radiation. In this work, a new and detailed calculation of the neutron flux for energies above 50 MeV is presented. This has been done by using state-of-the-art Monte Carlo astroparticle techniques and including real atmospheric profiles at each one of the next 23 exascale supercomputing facilities. Atmospheric impact in the flux and seasonal variations were observed and characterised, and the barometric coefficient for high-energy neutrons at each site was obtained. With these coefficients, potential risks of errors associated with the increase in the flux of energetic neutrons, such as the occurrence of single event upsets or transients, and the corresponding failure-in-time rates, can be anticipated just by using the atmospheric pressure before the assignation of resources to critical tasks at each exascale facility. For more clarity, examples about how the rate of failures is affected by the cosmic rays are included, so administrators will better anticipate which more or less restrictive actions could take for overcoming errors.

Auxetics and Other Systems with Unusual Characteristics

Auxetics and Other Systems with Unusual Characteristics