The design of sparse matrix storage formats is essential to achieve high-performance sparse kernels in modern parallel architectures. The bitmap-based bmSparse format was designed with the SPGEMM operation as its main focus, but it shows potential for other operations as well when the sparse matrix has a convenient structure. In this paper, we propose a new SPMV kernel that greatly improves the load balance of previous open-source implementations and utilizes the GPU resources more efficiently. The results show speedups of up to $100 \times$ regarding other SPMV kernels for bmSparse. Finally, we leverage a hybrid implementation between the new and existing approach, selecting the kernel which will probably result in a better performance depending on each matrix characteristics.
High-performance computing (HPC) data centers are key for advancing research and industry, but their high energy consumption and environmental impact pose critical sustainability challenges. Few HPC centers currently use renewable energy sources, particularly hydrogen-based storage, or integrate waste heat reuse. Recently, Digital Twins (DTs) are proving crucial for advancing sustainable HPC centers, enabling the integration of advanced monitoring and energy optimization with renewable energy sources. They simulate operations, predict maintenance needs, balance workloads, and support renewable integration for enhanced efficiency. The paper first reviews current practices, focusing on the potential of DTs to improve HPC sustainability by dynamically managing resources and optimizing systems, including cooling, power distribution, and load balancing. Then, it illustrates the University of Turin’s Sustainable HPC4AI (S-HPC4AI) project, which aims to develop a low impact HPC center to support Artificial Intelligence (AI) research across diverse scientific fields. The facility will feature renewable energy sources, advanced self-cooling, and waste heat recovery, setting a benchmark for energy-efficient, low-carbon HPC infrastructure. Central to the project is the integration of hydrogen and solar energy, with photovoltaic systems providing clean power and hydrogen fuel cells serving as reliable backup sources, reducing reliance on fossil fuels. Waste heat can be transformed into a productive resource for a local automated phenotyping system, reducing energy consumption and environmental impact on the overall. Furthermore, a DT will be developed, integrating BIM with sensors data about performance, resource usage and operating conditions, enabling real-time monitoring and predictive analytics for managing power, cooling, and energy resources. The potential and challenges of the S-HPC4AI model are discussed, suggesting possible solutions.
The project aims to develop innovative data management solutions (database and file system archive) for the Spoke 3 division of the Italian National Center for HPC, Big Data, and Quantum Computing, rigorously applying the FAIR principles and Open Science methodologies. This system provides a robust and scalable framework designed to address big data challenges in astrophysics, which requires customized data models and management solutions. It focuses on managing observations from the Gaia and Fermi space missions, as well as simulations generated using the Pluto and Ramses codes, with the goal of supporting requests from the astrophysical Italian community on particular scientific cases.For Gaia data, we identified the HDF5 as a suitable format to improve data access and analysis outside the data reduction pipelines. The HDF5 format allows the nesting of primitive fields, arrays, and complex objects, thereby enhancing accessibility and computational efficiency. For the data management of the FT1 and FT2 data products from the Fermi Observatory, a common data model has been chosen to provide a unique point of access to the scientific data, employing logic for database queries and offering access to data products over a long observing period. Finally, to facilitate user interaction, a custom web application has been developed, providing a user-friendly interface for accessing and querying the archived metadata, additional “Cut & Merge” and “Transits” services, and supporting also secure authentication methods to ensure data privacy and compliance. These comprehensive resources enhance data accessibility and usability through the Spoke3 Archive Infrastructure, empowering the scientific community to explore and analyze astronomical data more effectively.
Celestial objects are known to be change in brightness over time, driven by a diverse combination of physical processes, whose time scales range from sub-milliseconds to billions of years. Stingray is an open-source Python package that brings advanced time series analysis techniques to the astronomical community, with a focus on high-energy astrophysics, but built on top of general-purpose classes and methods that are designed to be easily adapted and extended to other use cases. We describe the work being done to adapt Stingray to the analysis of large data archives. In particular, we measure the performance and scalability of Stingray and use parallel computing to speed up selected parts of the code.
The C++ language continually evolves through formal specifications established by its standards committee, proposing new features to maintain $\mathrm{C}++$ as a relevant programming language while improving usability, performance, and portability across platforms. With the addition of parallel Standard Template Library (STL) algorithms in C++17, programmers can now leverage parallel processing capabilities via vendor-neutral parallel execution policies. This study presents an adaptation of the NAS Parallel Benchmarks (NPB)—a well-established suite of applications for evaluating parallel architectures-by porting its sequential C-style code to use C++ STL abstractions and performance-portable parallelism features. Our goals are to (1) assess the suitability of C++ STL for scientific applications like the ones in the NPB and (2) provide a comparative performance and portability of STL algorithms’ parallel execution policies across different multicore architectures (x86 and AArch64). Results indicate that the performance of parallel STL algorithms is often close to that of optimized handwritten versions (OpenMP, Intel TBB, and FastFlow) on different architectures, with notable shortfalls. Across all NPB benchmarks, the STL algorithms’ geometric mean shows sequential execution times that are between 3.76% and $\mathrm{6. 9 \%}$ higher, while parallel executions may reach a geometric mean of up to $\mathrm{2 1. 2 1 \%}$ higher execution time.
Modern Astrophysics and Cosmology (A&C) projects produce immense data volumes, necessitating advanced software tools for data access, storage, and analysis. Visualization Interface for the Virtual Observatory (VisIVO) is one such tool enabling multi-dimensional data analysis and knowledge discovery across complex astrophysical datasets. Leveraging containerization and virtualization, VisIVO has been deployed on various distributed computing platforms. Additionally, Blender, an open-source 3D suite, provides robust tools for rendering and processing volumetric data, making it suitable for visualizing complex datasets. At the SPACE Center of Excellence these tools are being adapted for high-performance visualization of cosmological simulations performed with GADGET and ChaNGa on pre-exascale systems. However, implementing high-performance visualization on diverse HPC platforms presents several challenges, including hardware and software compatibility, data management, scalability, performance portability, and efficient resource allocation. This paper outlines strategies to integrate VisIVO with workflow frameworks and streaming platforms to address these challenges. Workflow frameworks enhance portability, scheduling, and reproducibility of visualization workflows on pre-exascale systems used in A&C simulations. We also discuss the use of streaming platforms to enable concurrent (i.e. in-situ) analysis and visualization of simulations, reducing the need to store full simulation data by leveraging distributed databases that stream the output data in real time. Lastly, we present an adaptation of Blender to handle large-scale particle-based astrophysical data, offering high-quality visualization with interactive exploration capabilities.
High-performance computing (HPC) and data centers increasingly rely on Graphics Processing Units (GPUs) in large supercomputers. Yet, this reliance poses challenges in quickly understanding the performance impacts of code changes. This paper introduces NAV, a versatile tool designed to rapidly analyze and compare GPU performance traces. Built upon NVIDIA’s NsightTM Systems (NSYS), NAV enhances NSYS’s capabilities by accelerating visualization for large traces, increasing the variety of data representation formats, and adding comparative analysis capabilities. Our tool, NAV, provides complimentary functions on top of NSYS to quickly access performance data and perform comparative analysis. NAV offers detailed visual and written representations of traces at various granularity levels and efficiently handles large trace files through parallelization. It extracts trace data 1.15 to 3.5 times faster than comparable NSYS recipes for typical developer trace sizes, automating the generation of valuable data representations and streamlining workload analysis. This paper outlines NAV’s key features and functionalities, demonstrating its effectiveness through use cases that highlight its benefits for rapid application analysis and assessing the impact of code changes.
Multiple Sequence Alignment (MSA) is an important operation in Bioinformatics, used to simultaneously compare 3 or more sequences. The MSA problem was proven NP-Hard, so strategies have been proposed to reduce the search space and solve it in parallel. Recently, asymmetric multicore processors (AMPs) have become popular, with performance and energy-efficient cores, like the P-Cores and E-cores from Intel. However, parallel MSA applications have complex access patterns and adapting them for AMPs can be challenging. In this paper, we propose PA-Star21, an asymmetric-aware strategy based on A-Star, which computes optimal MSAs taking asymmetry into account when distributing the search space among threads. Our experimental results show that the proposed optimizations can reduce considerably the average execution time of PA-Star2 achieving a speedup of up to 7.70×. We also show that the asymmetric-aware strategy can reduce the average execution time for one of the hardest sequences set from the BAliBASE benchmark, when compared to the symmetric counterpart. Finally, we show that our approach is energy-efficient.11PA-Star2 is open source and the code is publicly available at1PA-Star2 is open source and the code is publicly available at https://github.com/danielsundfeld/astar_msa
Massive Open Online Courses (MOOCs) represent an accessible and user-friendly tool for disseminating innovative and cutting-edge topics to broad segments of civil society via online learning platforms, enabling users to learn at their own pace and on their own schedule. In this contribution, we describe the design and the implementation of a Massive Open Online Course on Parallel Computing and High-Performance Computing, developed for Federica Web Learning: the University Center for innovation, experimentation, and dissemination of multimedia teaching at the University of Naples Federico II.
The HaMMon project is the outcome of an industrial partnership that includes many Italian research institutions and private companies. It is led by UnipolSai and Leitha, and funded by the ICSC, the Italian National Research Center for High Performance Computing, Big Data and Quantum Computing.The ambition of HaMMon is to build a flexible and scalable platform to analyze the hydrogeological and atmospheric balance of the Italian territory. The project aims to expand the current knowledge in hazard mapping, monitoring, and forecasting from an industrial perspective by leveraging innovative technologies and the interdisciplinary activities carried out by the ICSC.In this work, we present the cloud-HPC infrastructure deployed in the High-Performance Computing for Artificial Intelligence (HPC4AI) green data center of the University of Turin which supports the testing and development of HaMMon’s applications and services. We describe the current activities and preliminary results related to the integration of Photogrammetry techniques, Data Visualization and Artificial Intelligence technologies, applied on aerial images, to assess extreme natural events and evaluate their impact on risk-exposed assets.
The Analog Ensemble (AnEn) method can be used to reconstruct incomplete time series using correlated series. Since the AnEn method may use data including several variables through long periods of time, its storage and computational cost may be substantial, slowing down reconstructions. This paper presents a full GPU implementation of the AnEn method, based on PyCUDA, that leads to a significant a cceleration of its execution. The implementation resorts to several techniques that seek to minimize the consumption of GPU global memory in the various steps of the AnEn algorithm, thus making room for larger input datasets. This is further reinforced by the use of batch processing as a way to automatically adapt the datasets size to the GPU memory available. The GPU implementation was tested on a meteorological dataset spanning 10 years, exhibiting a 30-fold speedup in the reconstruction time against a comparable CPUbased multicore version executed with up to 48 cores. The impact on the reconstruction error of changes on several important parameters of the implementation was also assessed.
Adversarial attacks are now becoming quite a dangerous means of disrupting image processing systems that use machine learning methods for decision making. Therefore, developing effective countermeasures against adversarial attacks is becoming quite an important area of cybersecurity. The paper proposes a noise-based approach to countering adversarial attacks that is augmented with neural-cleanse and jpeg-compression technologies. The idea of the proposed approach is that adding noise distorts the effect of an adversarial attack, and neural cleaning and jpeg compression eliminate the consequences of such an effect. The paper examines the three most well-known types of adversarial attacks: Fast Gradient Sign Method, Zeroth Order Optimization and One Pixel Attack. These attacks manipulate input data, resulting in misclassification or incorrect predictions by exploiting high-frequency components that are undetectable to humans. The research was carried out on two datasets: MNIST-JPG and PC Parts Images. Two types of noise were used: Gaussian and Poisson. During the experiments, optimal parameters for these types of noise were found, ensuring maximum accuracy of image recognition after exposure to adversarial attacks.
In recent years, multi-core systems with NonUniform Memory Access (NUMA) architectures are becoming prevalent. Multithreaded applications that execute on NUMA architectures play a pivotal role in getting better performance. The difficulties associated with multi-threaded applications are more pronounced because of compatibility issues with synchronization objects and the underlying NUMA architectures. Locks are fundamental synchronization primitives to avoid data races and achieve high performance in multi-core systems. With the increasing performance demands, locks have been refined and optimized with the architectural features of the NUMA systems. This paper proposes a novel synchronization object, namely CNS Lock (Compact NUMA-aware lock with a Standard interface) for NUMA systems. The CNS lock utilizes a single queue irrespective of the number of NUMA nodes and minimizes the number of atomic operations required for lock acquisition and release by prioritizing threads based on their NUMA node affinity. It considers the current thread’s NUMA node number to get the efficiency of a NUMA-aware design. CNS lock provides a standard interface to facilitate seamless integration into existing applications without knowing the internals. The CNS Lock strategically optimizes lock handovers and minimizes memory latency by employing a unified queue system across NUMA nodes. Our experimental results show that the CNS lock significantly outperforms the state-of-the-art locking mechanisms and makes it one of the essential tools of systems with NUMA architectures for better synchronization.
The solver module of the Astrometric Verification Unit-Global Sphere Reconstruction (AVU-GSR) pipeline aims to find the astrometric parameters of $\sim 10^{8}$ stars in the Milky Way, the attitude and instrumental settings of the Gaia satellite, and the parametrized post Newtonian parameter $\gamma$ with a resolution of 1 0 - 100 micro – arc seconds. To perform this task, the code, which runs in production on Leonardo CINECA infrastructure, solves a system of linear equations with the iterative LSQR algorithm, where the coefficient matrix is large (10-50 TB) and sparse and the iterations stop when least square convergence is reached. The solver was ported to GPU with CUDA, obtaining a $\sim 14 x$ acceleration factor over an original version CPU-parallelized with OpenMP. This work concentrates on a code section dedicated to covariances calculation, representing an important scientific task for Gaia mission, since the problems unknowns present strong correlations. Given the number of unknowns at mission end, the variances-covariances matrix is expected to occupy $\sim 1$ EB, which represents a substantial “Big Data” issue. To compute a subset of the total covariances, we defined an I/Obased pipeline made of two jobs. The first job, the LSQR, writes the files every $i \operatorname{tnCov} C P$ iterations, and the second job reads them and calculates the corresponding covariances. The two jobs can be launched either in sequence or concurrently. Previous studies demonstrated that the covariances calculation does not significantly slowdown the AVU-GSR production up to $\sim 3 \times 10^{7}$ covariances. Here we investigate the performance of the covariances pipeline as a function of $i t n \operatorname{Cov} C P$. The results show that writing smaller files more frequently or writing larger files less frequently does not affect the global performance of the solver, whose speed only depends on the number of covariances to calculate and of system unknowns.
In this work, we present a Deep Learning framework to predict the progenitor star’s characteristics of Supernovae (SNe) from their observed light curves. This task is crucial for astrophysics, as it can provide insights into the evolution of the star before the explosion and into the SN explosion mechanism. In fact, there is no direct mapping between the observed light curves and the progenitor’s characteristics, and the common techniques used to infer them are indirect, meaning that they rely on the comparison between the observed light curves and the light curves generated with some physical model simulation using the supposed progenitor’s characteristics as input. However, the physical models used to generate the light curves are not perfect, being either computationally expensive or based on simplifying assumptions. Here, we train a machine learning model on a dataset of light curves generated with a semi-analytical model — which is computationally efficient and accurate for the problem at hand — to predict the progenitor’s mass, radius, energy, and nickel mass from the observed light curves of SNe similar to SN 1987A. Our results show that the Deep Learning model effectively learns the complex mapping between the observed light curves and the progenitor’s characteristics with a low mean absolute percentage error, and we note that the proposed framework is general and can be applied to other types of SNe without significant modifications.
Early anomaly detection in automotive systems is crucial for enhancing user safety and enabling timely corrective actions, thereby minimizing the risks associated with system malfunctions. This paper presents an approach for implementing Artificial Intelligence (AI)-based algorithms for anomaly detection in the automotive domain, leveraging the RISC-V architecture in conjunction with Domain-Specific Accelerators (DSAs). By exploiting the efficiency of DSAs, the proposed system aims to achieve faster anomaly detection compared to traditional processing methods. A detailed comparison is conducted between the performance of executing the AI-based anomaly detection algorithm on the RISC-V core versus offloading it to an optimized hardware accelerator tailored to the specific AI model. The goal of this work is to provide valuable insights into the potential of RISCV and DSAs to enhance AI-driven safety mechanisms, contributing to the development of more reliable automotive systems.
Current parallel systems are increasingly heterogeneous, mixing devices of different types and computing capabilities. Exploiting multiple different devices for the same application continues to be a challenge that ranges from technical problems related to synchronizing and communicating diverse devices to problems of load distribution and flexibility to adjust the computation to the platform resources. In this work, we study the problem of using and extending a heterogeneous portability layer to program and adapt HSOpticalFlow to heterogeneous platforms. HSOpticalFlow is a streaming application to estimate the apparent movement of objects in a sequence of images. It is a simple but characteristic example of the structure of applications based on multilevel ILS (Iterative Loop Stencil), also known as multi-grid methods, applied to a sequence of inputs. Starting from the original CUDA reference code, we present a methodology and programming techniques based on the Controller programming model to implement it as a pipeline among multiple devices. We discuss a technique to determine a proper work partition and mapping for a set of devices. This allows for building very efficient parallel solutions, using similar devices or taking advantage of devices with lower computing power, to reduce the load and increase the productivity of more powerful ones. We present the results of an experimental study using several GPUs of different vendors, architectures, and generations, showing that this solution allows combinations of devices to be efficiently exploited to improve performance. Specifically, the results include speedups of 1.91x using two NVIDIA A100 GPUs and 1.21x using one NVIDIA V100 GPU and one AMD WX9100 GPU, which is about $3 x$ slower than the NVIDIA GPU for this application.
This paper introduces a web-based tool designed to organise and present visual representations of performance, portability, and productivity (P3) data from previously published scientific studies. The P3 Explorer operates as both an open repository of application performance data and a data dashboard, providing visual heuristic analyses of performance portability and developer productivity, created using Intel’s P3 Analysis library. The goal of the project is to create a community-led database of P3 studies to better inform application developers of alternative approaches to developing new applications that target high performance on diverse hardware, considering the productivity of the developer. In this paper, we evaluate our tool using a recently published study outlining a performance portable domain-specific language for particle-in-cell applications.
The Vlasov-Poisson systems of equations (VP) describes the evolution of a distribution of collisionless particles under the effect of a collective-field potential. VP is at the basis of the study of the gravitational instability of cosmological density perturbations in Dark-Matter (DM), but its range of application extends to other fields, such as plasma physics. In the case of Cold Dark Matter, a single velocity is associated with each fluid-element (or particle), the initial condition presents a stiff discontinuity. This creates problems such as diffusion or negative distribution function when a grid based method is used to solve VP. In this work we want to highlight this problem, focusing on the technical aspects of this phenomenon. By comparing different finite volume methods and a spectral method we observe that, while all integration schemes preserve the invariants of the system (e.g, energy), the physical observable of interest, i.e., the density, is not correctly reproduced. We thus compare the density obtained with the different Eulerian integration schemes with the result obtained from a reference N -body method. We point out that the most suitable method to solve the VP system for a self-gravitating system is a spectral method.
HPC is a widely used term, often referred to the applications, architectures and programming models and tools targeting highly parallel machines such as those of the top500.org lists. Recent advances in computing hardware resources require application of HPC techniques when using much smaller machines. Indeed, proper parallel programming tools and applications are needed also to exploit parallel hardware resources in personal computers (laptops, desktops, servers). This paper outlines key challenges in designing master’s degree programs in HPC and shares lessons learned from various experiences in developing and implementing such programs in Italy and Europe.
