Search

High-Performance Computing can be accelerated using long-vector architectures. However, creating efficient coding implementations for these architectures can be challenging. This Master's thesis focuses on implementing four well-known and widely-used graph processing algorithms using the RISC-V Vector Extension, leveraging an experimental system in an FPGA. I present a graph storage format that benefits from long vectors and describe how these four algorithms can be rewritten to utilize it. This thesis also introduces an instrumentation tool for FPGA that I developed to link the output of electrical engineering software with performance analysis tools for HPC. This tool allows users to visualize information coming from the logic analyzer internal to the FPGA with powerful visualization tools, permitting fine-grain analysis of the FPGA signals correlated with the code running on it. This tool has been integrated into the experimental performance analysis tools of BSC. In this thesis I leverage this tool to analyze and improve my implementations of graph algorithms for long-vector architectures, collecting the process and thoughts behind each optimization. Finally, I compare the performance of my vector implementations with other machines, such as the NEC SX-Aurora, a commercial RISC-V board, and an Intel chip.

En el món del "High-Performance Computing", el processador és essencial. Recentment, Europa està fent grans esforços en promoure tecnologia europea. La "European Processor Initiative" sorgeix d'aquest esforç. Com a part de la iniciativa, múltiples processadors estan sent dissenyats. Alguns implementant l'arquitectura "RISC-V", una "ISA" "open-source". Al llarg del desenvolupament del processador, disposar d'eines és fonamental per facilitar el testatge i automatitzar tasques. Aquest treball final de grau es focalitza en millorar una "pipeline" de "Continuous integration" emprada per detectar errors en un entorn Linux i en una "Field Programmable Gate Array (FPGA)" emulant un comportament d'usuari final. In the High-Performance Computing world, the processor is essential. In recent years, Europe has devoted a lot of effort into promoting European technology. The European Processor Initiative stems from this effort. As part of the initiative, multiple processors are being developed. Some implementing the RISC-V architecture, an open-source ISA. During the development of a processor, tools are fundamental to ease testing and automatize tasks. This final degree project focuses on improving a Continuos Integration pipeline used to detect bugs in an Field Programmable Gate Array (FPGA) and Linux environments emulating final user behaviour.

The landscape of High Performance Computing (HPC) system architectures keeps expanding with new technologies and increased complexity. To improve the efficiency of next-generation compute devices, architects are looking for solutions beyond the commodity CPU approach. In 2021, the five most powerful supercomputers in the world use either GP-GPU (General-purpose computing on graphics processing units) accelerators or a customized CPU specially designed to target HPC applications. This trend is only expected to grow in the next years motivated by the compute demands of science and industry. As architectures evolve, the ecosystem of tools and applications must follow. The choices in the number of cores in a socket, the floating point-units per core and the bandwidth through the memory hierarchy among others, have a large impact in the power consumption and compute capabilities of the devices. To balance CPU and accelerators, designers require accurate tools for analyzing and predicting the impact of new architectural features on the performance of complex scientific applications at scale. In such a large design space, capturing and modeling with simulators the complex interactions between the system software and hardware components is a defying challenge. Moreover, applications must be able to exploit those designs with aggressive compute capabilities and memory bandwidth configurations. Algorithms and data structures will need to be redesigned accordingly to expose a high degree of data-level parallelism allowing them to scale in large systems. Therefore, next-generation computing devices will be the result of a co-design effort in hardware and applications supported by advanced simulation tools. In this thesis, we focus our work on the co-design of scientific applications and long vector architectures. We significantly extend a multi-scale simulation toolchain enabling accurate performance and power estimations of large-scale HPC systems. Through simulation, we explore the large design space in current HPC trends over a wide range of applications. We extract speedup and energy consumption figures analyzing the trade-offs and optimal configurations for each of the applications. We describe in detail the optimization process of two challenging applications on real vector accelerators, achieving outstanding operation performance and full memory bandwidth utilization. Overall, we provide evidence-based architectural and programming recommendations that will serve as hardware and software co-design guidelines for the next generation of specialized compute devices. El panorama de las arquitecturas de los sistemas para la Computación de Alto Rendimiento (HPC, de sus siglas en inglés) sigue expandiéndose con nuevas tecnologías y complejidad adicional. Para mejorar la eficiencia de la próxima generación de dispositivos de computación, los arquitectos están buscando soluciones más allá de las CPUs. En 2021, los cinco supercomputadores más potentes del mundo utilizan aceleradores gráficos aplicados a propósito general (GP-GPU, de sus siglas en inglés) o CPUs diseñadas especialmente para aplicaciones HPC. En los próximos años, se espera que esta tendencia siga creciendo motivada por las demandas de más potencia de computación de la ciencia y la industria. A medida que las arquitecturas evolucionan, el ecosistema de herramientas y aplicaciones les debe seguir. Las decisiones eligiendo el número de núcleos por zócalo, las unidades de coma flotante por núcleo y el ancho de banda a través de la jerarquía de memoría entre otros, tienen un gran impacto en el consumo de energía y las capacidades de cómputo de los dispositivos. Para equilibrar las CPUs y los aceleradores, los diseñadores deben utilizar herramientas precisas para analizar y predecir el impacto de nuevas características de la arquitectura en el rendimiento de complejas aplicaciones científicas a gran escala. Dado semejante espacio de diseño, capturar y modelar con simuladores las complejas interacciones entre el software de sistema y los componentes de hardware es un reto desafiante. Además, las aplicaciones deben ser capaces de explotar tales diseños con agresivas capacidades de cómputo y ancho de banda de memoria. Los algoritmos y estructuras de datos deberán ser rediseñadas para exponer un alto grado de paralelismo de datos permitiendo así escalarlos en grandes sistemas. Por lo tanto, la siguiente generación de dispósitivos de cálculo será el resultado de un esfuerzo de codiseño tanto en hardware como en aplicaciones y soportado por avanzadas herramientas de simulación. En esta tesis, centramos nuestro trabajo en el codiseño de aplicaciones científicas y arquitecturas vectoriales largas. Extendemos significativamente una serie de herramientas para la simulación multiescala permitiendo así obtener estimaciones de rendimiento y potencia de sistemas HPC de gran escala. A través de simulaciones, exploramos el gran espacio de diseño de las tendencias actuales en HPC sobre un amplio rango de aplicaciones. Extraemos datos sobre la mejora y el consumo energético analizando las contrapartidas y las configuraciones óptimas para cada una de las aplicaciones. Describimos en detalle el proceso de optimización de dos aplicaciones en aceleradores vectoriales, obteniendo un rendimiento extraordinario a nivel de operaciones y completa utilización del ancho de memoria disponible. Con todo, ofrecemos recomendaciones empíricas a nivel de arquitectura y programación que servirán como instrucciones para diseñar mejor hardware y software para la siguiente generación de dispositivos de cálculo especializados. Arquitectura de computadors

Acceleration is a widely used technique in the field of High-Performance Computing (HPC) to improve the performance of standard CPU powered computational nodes. Examples of widely used accelerators in HPC are graphic cards (\textit{e.g.}, Nvidia GP-GPU) and vector accelerators (\textit{e.g.}, NEC Aurora-SX). Custom acceleration is often achieved using FPGAs. One of the challenges of adopting FPGA-based accelerators in HPC is their programmability. OmpSs@FPGA is a programming model developed at the Barcelona Supercomputing Center (BSC) that allows adding FPGA acceleration to an application in an easy and device-independent way. In this project, we used OmpSs@FPGA to accelerate in hardware the generation of random numbers. Monte Carlo algorithms are widely used in scientific applications in the context of HPC. These algorithms are characterized by its need of random number input. It is possible to offload random number generation using hardware acceleration techniques. We implemented three different versions of a linear-feedback shift register (LFSR) random number generator: a CPU-only version and two OmpSs@FPGA accelerated versions. The performance of the three implementations has been evaluated in the context of the Ising model with the Metropolis algorithm, which belongs to the family of Monte Carlo methods used in statistical physics. This provides a real world use case that makes heavy use of pseudorandom number generation. While we verified that the Ising model is indeed limited by the generation of random numbers, our results show that the acceleration of random number generation using FPGA needs to be carefully deployed. Our preliminary results show a high dependency between the achieved performance and the overhead introduced by the offloading infrastrcture. L'acceleració hardware és una tècnica àmpliament utilitzada en el camp de la computació d'alt rendiment (HPC) per augmentar el rendiment dels nodes de càlcul estàndard basats en CPU. Alguns exemples d'acceleradors d'ús comú en HPC poden ser les targetes gràfiques (p. ex. Nvidia GP-GPU) o els acceleradors vectorials (p. ex. NEC Aurora-SX). L'acceleració especialitzada es pot obtenir amb FPGAs. Una de les dificultats que limiten l'ús dels acceleradors FPGA en el món del HPC és la seva programabilitat. OmpSs@FPGA és un model de programació desenvolupat al Barcelona Supercomputing Center (BSC) que permet incorporar acceleració FPGA a una aplicació de manera fàcil. En aquest projecte, hem utilitzat OmpSs@FPGA per accelerar en hardware la generació de nombres aleatoris. Els algoritmes Monte Carlo s'utilitzen àmpliament en aplicacions científiques i HPC. Es caracteritzen per la necessitat de nombres aleatoris per al seu funcionament. És possible aplicar offloading a la generació de nombres aleatoris mitjançant acceleració hardware. Hem implementat tres versions d'un generador de nombres aleatoris basat en LFSR: una versió CPU-only i dues versions accelerades amb OmpSs@FPGA. El rendiment de les tres implementacions s'ha avaluat en el context del model d'Ising amb l'algoritme de Metropolis, que pertany a la família dels algoritmes Monte Carlo utilitzats en el camp de la física estadística. Això ens proporciona un cas d'ús real que en fa un ús intensiu de la generació de nombres aleatoris. Tot i haver verificat que el model d'Ising queda limitat per la generació de nombres aleatoris, els resultats obtinguts mostren que la generació de nombres aleatoris accelerada amb FPGA s'ha d'implementar amb cura. Els nostres resultats preliminars presenten una forta dependència entre el rendiment obtingut i l'overhead introduït per la infraestructura d'offloading

Este trabajo presenta una optimización del cálculo de la Transformada de Fou- rier para arquitecturas con vectores largos. Al ser un cálculo muy presente en una gran variedad de códigos científicos, su optimización siempre es asunto de actualidad e investigación. En el documento encontramos un estudio exhaustivo del algoritmo, explorando diferentes posibles optimizaciones. Finalmente se expone un algoritmo para cálcular la Transformada de Fourier que se aprovecha de los beneficios de las arquitecturas vectoriales y consigue un mejor rendimiento que las librerías ya exis- tentes en ciertos casos. In this thesis, an optimization for the computing of the Fourier Transform is presented for architectures with long vectors. Since this calculation has a broad presence in scientific codes, its optimization is always a relevant and open topic for investigation. In this document, we find an extensive study on the algorithm which explores different ways of optimization. Finally, an algorithm is proposed to compute the Fourier Transform taking advantage of a vector architecture, achieving better performance than the existing libraries for a set of cases.

In this thesis, I present a methodology to evaluate High-Performance Computing systems. The method relies on measurements at three levels: architectural features through micro-benchmarks; system software and tools through benchmarks and analysis of the code generated by the compiler; and sustained performance though scientific applications. I apply the method to three state-of-the-art High-Performance Computing clusters deployed at the Barcelona Supercomputing Center.

Container technologies usage has been growing in data centers and supercomputer facilities during the last years. More and more people are benefiting from the portability, flexibility and other traits of containers. Despite being some evaluation performed in the field of HPC, there is still a lot of research to be done on the matter. Also, the system-wide deployment and usage of a container infrastructure in an HPC system requires a non-negligible effort. Therefore a part of the study is also to analyze if the deployment effort is worth it. In this thesis I study the performance and energy consumption of Singularity containers with HPC workloads of real scientific applications (OpenFOAM, GROMACS) on emerging HPC architectures (Armv8). In addition, I make some suggestions on the Singularity deployment from a user and a System Administrator perspective. Results show almost no performance nor energy differences between the executions of applications ran inside and outside of a container when tested on a small number of compute nodes. Containerization with Singularity seems overall a good tool to have in an HPC system.

This project is about porting and tuning the Mont-Blanc benchmarks to the multicore ARM 64 bits architecture. The Mont-Blanc benchmarks are part of the Mont-Blanc European project and they have been developed internally in the BSC (Barcelona Supercomputing Center). The project will explore the possibilities that an ARM architecture can offer running in a HPC (High Performance Computing) setup, this includes to learn how to tune and adapt a parallelized computer program and analyze its execution behavior. As part of the project, we will analyze the performance of each benchmark using instrumentation tools such like Extrae and Paraver. Each benchmark will be adapted, tuned and executed mainly in the three new Mont-Blanc mini-clusters, Thunder (ARMv8 custom), Merlin (ARMv8 custom) and Jetson TX (ARMv8 cortex-a57) using the OmpSs programming model. The evolution of the performance obtained will be shown followed by a brief analysis of the results after each optimization. Aquest projecte es basa en adaptar i afinar els Mont-Blanc benchmarks a l’arquitectura multinucli ARM 64 bits. Els Mont-Blanc benchmarks formen part del projecte Europeu Mont-Blanc i han estat desenvolupats internament en el BSC (Barcelona Supercomputing Center). Aquest projecte explorarà el potencial d’usar l’arquitectura ARM en un entorn HPC (High Performance Computing), això inclou aprendre a adaptar i afinar un programa paral·lel, i analitzar el seu comportament durant l’execució. Com a part del projecte, s’analitzarà el rendiment de cada benchmark usant eines d’instrumentació com Extrae o Paraver. Cada benchmark serà adaptat, afinat i executat en els tres nous miniclústers de Mont-Blanc, Thunder (ARMv8 personalitzat), Merlin (ARMv8 personalitzat) i Jetson TX (ARMv8 cortex-a57) usant el model de programació OmpSs. Es mostrarà l’evolució del rendiment, seguit d’una breu explicació dels resultats després de cada optimització. Este proyecto se basa en adaptar y afinar los Mont-blanc benchmarks a la arquitectura multi-núcleo ARM 64 bits. Los Mont-Blanc benchmarks forman parte del proyecto Europeo Mont-Blanc y han sido desarrollados internamente en el BSC (Barcelona Supercomputing Center). Este proyecto explorará el potencial de usar la arquitectura ARM en un entorno HPC (High Performance Computing), esto incluye aprender a adaptar y afinar un programa paralelo, y analizar su comportamiento durante la ejecución. Como parte del proyecto, se analizará el rendimiento de cada benchmark usando herramientas de instrumentación como Extrae o Paraver. Cada benchmark será adaptado, afinado y ejecutado en los tres nuevos mini-clústeres de Mont-Blanc, Thunder (ARMv8 personalizado), Merlin (ARMv8 personalizado) y Jetson TX (ARMv8 cortex-a57) usando el modelo de programación OmpSs. Se mostrará la evolución del rendimiento obtenido, y una breve explicación de los resultados después de cada optimización.

High Performance Computing (HPC) is used for running advanced application programs efficiently, reliably, and quickly. In earlier decades, performance analysis of HPC applications was evaluated based on speed, scalability of threads, memory hierarchy. Now, it is essential to consider the energy or the power consumed by the system while executing an application. In fact, the High Power Consumption (HPC) is one of biggest problems for the High Performance Computing (HPC) community and one of the major obstacles for exascale systems design. The new generations of HPC systems intend to achieve exaflop performances and will demand even more energy to processing and cooling. Nowadays, the growth of HPC systems is limited by energy issues Recently, many research centers have focused the attention on doing an automatic tuning of HPC applications which require a wide study of HPC applications in terms of power efficiency. In this context, this paper aims to propose the study of an oceanographic application, named OceanVar, that implements Domain Decomposition based 4D Variational model (DD-4DVar), one of the most commonly used HPC applications, going to evaluate not only the classic aspects of performance but also aspects related to power efficiency in different case of studies. These work were realized at Bsc (Barcelona Supercomputing Center), Spain within the Mont-Blanc project, performing the test first on HCA server with Intel technology and then on a mini-cluster Thunder with ARM technology. In this work of thesis it was initially explained the concept of assimilation date, the context in which it is developed, and a brief description of the mathematical model 4DVAR. After this problem’s close examination, it was performed a porting from Matlab description of the problem of data-assimilation to its sequential version in C language. Secondly, after identifying the most onerous computational kernels in order of time, it has been developed a parallel version of the application with a parallel multiprocessor programming style, using the MPI (Message Passing Interface) protocol. The experiments results, in terms of performance, have shown that, in the case of running on HCA server, an Intel architecture, values of efficiency of the two most onerous functions obtained, growing the number of process, are approximately equal to 80%. In the case of running on ARM architecture, specifically on Thunder mini-cluster, instead, the trend obtained is labeled as "SuperLinear Speedup" and, in our case, it can be explained by a more efficient use of resources (cache memory access) compared with the sequential case. In the second part of this paper was presented an analysis of the some issues of this application that has impact in the energy efficiency. After a brief discussion about the energy consumption characteristics of the Thunder chip in technological landscape, through the use of a power consumption detector, the Yokogawa Power Meter, values of energy consumption of mini-cluster Thunder were evaluated in order to determine an overview on the power-to-solution of this application to use as the basic standard for successive analysis with other parallel styles. Finally, a comprehensive performance evaluation, targeted to estimate the goodness of MPI parallelization, is conducted using a suitable performance tool named Paraver, developed by BSC. Paraver is such a performance analysis and visualisation tool which can be used to analyse MPI, threaded or mixed mode programmes and represents the key to perform a parallel profiling and to optimise the code for High Performance Computing. A set of graphical representation of these statistics make it easy for a developer to identify performance problems. Some of the problems that can be easily identified are load imbalanced decompositions, excessive communication overheads and poor average floating operations per second achieved. Paraver can also report statistics based on hardware counters, which are provided by the underlying hardware. This project aimed to use Paraver configuration files to allow certain metrics to be analysed for this application. To explain in some way the performance trend obtained in the case of analysis on the mini-cluster Thunder, the tracks were extracted from various case of studies and the results achieved is what expected, that is a drastic drop of cache misses by the case ppn (process per node) = 1 to case ppn = 16. This in some way explains a more efficient use of cluster resources with an increase of the number of processes.

The thesis evaluates the current state-of-the-art of RISC architectures in HPC. Studying the performance, power, and energy to solution in heterogeneous SoCs. For the evaluation 2 arm platforms (CPU+GPU, CPU+FPGA), 1 RISC-V platform and 1 Open Source RISC-V core running in an FPGA have been tested.