Context Of High Performance Computing Research Articles

Graph and convolutional neural network coupling with a high-performance large-eddy simulation solver

Computational Fluid Dynamics (CFD) traditionally relies on long-standing numerical simulation strategies for the Navier–Stokes equations. Recently, interest in data-driven hybrid CFD solvers has spiked, leveraging pre-computed datasets to enhance various weak links inside existing solvers, such as closure models, under-resolved physics, or even to guide numerical resolution strategies. Running these hybrid solvers, notably in High Performance Computing (HPC) environments, presents specific challenges. In particular, context-aware deep learning (e.g. Convolutional (CNN) and Graph (GNN) Neural Networks) is promising for this task, but requires passing data representations between the physics solver and the neural network. In relevant industrial configurations, CFD meshes can be Cartesian but highly irregular, or unstructured, both of which do not match the pixel/voxel structure needed to run CNNs. In addition, discrepancies in programming language and libraries are common between CFD and machine learning applications. This work explores the many challenges of running a parallel hybrid solver in an HPC context, through the coupling of the AVBP CFD solver with neural networks in turbulent combustion and wall friction modeling applications. The knowledge gained is showcased in this article, as well as assembled in an actionable open-source library.

COCKTAIL: Multicore Co-Optimization Framework With Proactive Reliability Management

High performance computing (HPC) servers aim to meet an increase in the number and complexity of tasks and, consequently, to address the energy efficiency challenge. In addition to energy efficiency, it is essential to manage lifetime limitations of power-hungry components of servers (e.g., cores and cache), hence avoiding server failure before its lifetime period. Traditional approaches focus on either using hybrid caches to reduce the leakage power of traditional static random-access memory (SRAM) cache, and thus increase the energy efficiency, or the trade-off between the lifetime and performance of multi-core processors. However, these approaches fall short in terms of flexibility and applicability for HPC tasks in terms of multi-parametric optimization including quality-of-service (QoS), lifetime reliability, and energy efficiency. As a result, in this paper we propose COCKTAIL, a holistic strategy framework to jointly optimize the energy efficiency of multi-core server processors and tasks performance in the HPC context, while guaranteeing the lifetime reliability. First, we analyze the best cache technology among traditional SRAM and resistive random access memory (RRAM), within the context of hybrid cache architectures, to improve the energy efficiency and manage cache endurance limits with respect to tasks requirements. Second, we introduce a novel efficient proactive queue optimization policy to reorder HPC tasks for execution considering their end time and possible reliability effects on the use of the hybrid caches. Third, we present a dynamic model predictive control (MPC)-based reliability management method to maximize task performance, by controlling the frequency, temperature, and target lifetime of the server processor. Our results demonstrate that, while consuming similar energy, COCKTAIL provides up to 60% QoS improvement when compared to latest state-of-the-art energy optimization and reliability management techniques in the HPC context. Moreover, our strategy guarantees a design lifetime longer than 5 years for the whole HPC system.

Machine Learning to Design an Auto-tuning System for the Best Compressed Format Detection for Parallel Sparse Computations

Many applications in scientific computing process very large sparse matrices on parallel architectures. The presented work in this paper is a part of a project where our general aim is to develop an auto-tuner system for the selection of the best matrix compression format in the context of high-performance computing. The target smart system can automatically select the best compression format for a given sparse matrix, a numerical method processing this matrix, a parallel programming model and a target architecture. Hence, this paper describes the design and implementation of the proposed concept. We consider a case study consisting of a numerical method reduced to the sparse matrix vector product (SpMV), some compression formats, the data parallel as a programming model and, a distributed multi-core platform as a target architecture. This study allows extracting a set of important novel metrics and parameters which are relative to the considered programming model. Our metrics are used as input to a machine-learning algorithm to predict the best matrix compression format. An experimental study targeting a distributed multi-core platform and processing random and real-world matrices shows that our system can improve in average up to 7% the accuracy of the machine learning.

Understanding the Determinants and Future Challenges of Cloud Computing Adoption for High Performance Computing

High performance computing (HPC) is widely recognized as a key enabling technology for advancing scientific progress, industrial competitiveness, national and regional security, and the quality of human life. Notwithstanding this contribution, the large upfront investment and technical expertise required has limited the adoption of HPC to large organizations, government bodies, and third level institutions. Recent advances in cloud computing and telecommunications have the potential to overcome the historical issues associated with HPC through increased flexibility and efficiency, and reduced capital and operational expenditure. This study seeks to advance the literature on technology adoption and assimilation in the under-examined HPC context through a mixed methods approach. Firstly, the determinants of cloud computing adoption for HPC are examined through a survey of 121 HPC decision makers worldwide. Secondly, a modified Delphi method was conducted with 13 experts to identify and prioritize critical issues in the adoption of cloud computing for HPC. Results from the quantitative phase suggest that only organizational and human factors significantly influence cloud computing adoption decisions for HPC. While security was not identified as a significant influencer in adoption decisions, qualitative research findings suggest that data privacy and security issues are an immediate and long-term concern.

Performance and Energy Footprint Assessment of FPGAs and GPUs on HPC Systems Using Astrophysics Application

New challenges in Astronomy and Astrophysics (AA) are urging the need for many exceptionally computationally intensive simulations. “Exascale” (and beyond) computational facilities are mandatory to address the size of theoretical problems and data coming from the new generation of observational facilities in AA. Currently, the High-Performance Computing (HPC) sector is undergoing a profound phase of innovation, in which the primary challenge to the achievement of the “Exascale” is the power consumption. The goal of this work is to give some insights about performance and energy footprint of contemporary architectures for a real astrophysical application in an HPC context. We use a state-of-the-art N-body application that we re-engineered and optimized to exploit the heterogeneous underlying hardware fully. We quantitatively evaluate the impact of computation on energy consumption when running on four different platforms. Two of them represent the current HPC systems (Intel-based and equipped with NVIDIA GPUs), one is a micro-cluster based on ARM-MPSoC, and one is a “prototype towards Exascale” equipped with ARM-MPSoCs tightly coupled with FPGAs. We investigate the behavior of the different devices where the high-end GPUs excel in terms of time-to-solution while MPSoC-FPGA systems outperform GPUs in power consumption. Our experience reveals that considering FPGAs for computationally intensive application seems very promising, as their performance is improving to meet the requirements of scientific applications. This work can be a reference for future platform development for astrophysics applications where computationally intensive calculations are required.

Resilient gossip-inspired all-reduce algorithms for high-performance computing: Potential, limitations, and open questions

We investigate the usefulness of gossip-based reduction algorithms in a high-performance computing (HPC) context. We compare them to state-of-the-art deterministic parallel reduction algorithms in terms of fault tolerance and resilience against silent data corruption (SDC) as well as in terms of performance and scalability. New gossip-based reduction algorithms are proposed, which significantly improve the state-of-the-art in terms of resilience against SDC. Moreover, a new gossip-inspired reduction algorithm is proposed, which promises a much more competitive runtime performance in an HPC context than classical gossip-based algorithms, in particular for low accuracy requirements.

Mixing HTC and HPC Workloads with HTCondor and Slurm

Traditionally, the RHIC/ATLAS Computing Facility (RACF) at Brookhaven National Laboratory (BNL) has only maintained High Throughput Computing (HTC) resources for our HEP/NP user community. We’ve been using HTCondor as our batch system for many years, as this software is particularly well suited for managing HTC processor farm resources. Recently, the RACF has also begun to design/administrate some High Performance Computing (HPC) systems for a multidisciplinary user community at BNL. In this paper, we’ll discuss our experiences using HTCondor and Slurm in an HPC context, and our facility’s attempts to allow our HTC and HPC processing farms/clusters to make opportunistic use of each other’s computing resources.

Evaluation of DVFS techniques on modern HPC processors and accelerators for energy‐aware applications

SummaryEnergy efficiency is becoming increasingly important for computing systems, in particular for large scale High Performance Computing (HPC) facilities. In this work, we evaluate, from a user perspective, the use of Dynamic Voltage and Frequency Scaling techniques, assisted by the power and energy monitoring capabilities of modern processors to tune applications for energy efficiency. We run selected kernels and a full HPC application on 2 high‐end processors widely used in the HPC context, namely, an NVIDIA K80 GPU and an Intel Haswell CPU. We evaluate the available trade‐offs between energy‐to‐solution and time‐to‐solution, attempting a function‐by‐function frequency tuning. We finally estimate the benefits obtainable running the full code on an HPC multi‐GPU node, with respect to default clock frequency governors. We instrument our code to accurately monitor power consumption and execution time without the need of any additional hardware, and we enable it to change CPUs and GPUs clock frequencies while running. We analyze our results on the different architectures using a simple energy‐performance model and derive a number of energy saving strategies, which can be easily adopted on recent high‐end HPC systems for generic applications.

A Performance Evaluation Method for Climate Coupled Models

Abstract In the High Performance Computing context, the performance evaluation of a parallel algorithm is carried out mainly by considering the elapsed time for running the parallel application with both different number of cores and different problem sizes (for scaled speedup). Typically, parallel applications embed mechanisms to efficiently use the allocated resources, guaranteeing for example a good load balancing and reducing the parallel overhead. Unfortunately, this assumption is not true for coupled models. These models were born from the coupling of stand-alone climate models. The component models are developed independently from each other and they follow different development roadmaps. Moreover, they are characterized by different levels of parallelization as well as different requirements in terms of workload and they have their own scalability curve. Considering a coupled model as a single parallel application, we can note the lacking of a policy useful to balance the computational load on the available resources. This work tries to address the issues related to the performance evaluation of a coupled model as well as answering the following questions: once a given number of processors has been allocated for the whole coupled model, how does the run have to be configured in order to balance the workload? How many processors must be assigned to each of the component models? The methodology here described has been applied to evaluate the scalability of the CMCC-MED coupled model designed by the ANS Division of the CMCC. The evaluation has been carried out on two different computational architectures: a scalar cluster, based on IBM Power6 processors, and a vector cluster, based on NEC-SX9 processors.

A Performance Evaluation Method for Coupled Models

In the High-Performance Computing context, the performance evaluation of a parallel algorithm is made mainly considering the elapsed time running the parallel application with both different number of cores or different problem sizes (for scaled speed-up). Typically, parallel applications embed mechanisms for efficiently using the allocated resources, guarantying for example a good load balancing and reducing the parallel overhead. Unfortunately, this assumption is not true for coupled models.These models are born from the coupling of stand-alone climate applications. The component models are developed independently from each other and they follow different development roadmaps. Moreover, they are characterized by different levels of parallelization, different requirements in terms of workload and they have their own scalability curve. Considering a coupled model as a single parallel application, we can note the lacking of a policy for balancing the computational load on the available resources. This work tries to address the issues related to performance evaluation of a coupled model, and to answer to the following questions: allocated a given number of processors for the whole coupled model, how to configure the run in order to balance the workload? How many processors must be assigned to each of the component models? The methodology here described has been applied for evaluating the scalability of the CMCC-MED coupled model designed by INGV and the ANS Division of the CMCC. The evaluation has been carried out on two different computational architectures: a scalar cluster based on IBM Power6 processors; and a vector cluster based on NEC-SX9 processors.

System-level virtualization research at Oak Ridge National Laboratory

System-level virtualization is today enjoying a rebirth as a technique to effectively share what had been considered large computing resources which subsequently faded from the spotlight as individual workstations gained in popularity with a “one machine–one user” approach. One reason for this resurgence is that the simple workstation has grown in capability to rival anything similar, available in the past. Thus, computing centers are again looking at the price/performance benefit of sharing that single computing box via server consolidation. However, industry is only concentrating on the benefits of using virtualization for server consolidation (enterprise computing) whereas our interest is in leveraging virtualization to advance high-performance computing (HPC). While these two interests may appear to be orthogonal, one consolidating multiple applications and users on a single machine while the other requires all the power from many machines to be dedicated solely to its purpose, we propose that virtualization does provide attractive capabilities that may be exploited to the benefit of HPC interests. This does raise the two fundamental questions: is the concept of virtualization (a machine “sharing” technology) really suitable for HPC and if so, how does one go about leveraging these virtualization capabilities for the benefit of HPC. To address these questions, this document presents ongoing studies on the usage of system-level virtualization in a HPC context. These studies include an analysis of the benefits of system-level virtualization for HPC, a presentation of research efforts based on virtualization for system availability, and a presentation of research efforts for the management of virtual systems. The basis for this document was the material presented by Stephen L. Scott at the Collaborative and Grid Computing Technologies meeting held in Cancun, Mexico on April 12–14, 2007.

High Performance Computing Productivity Model Synthesis

The Defense Advanced Research Projects Agency (DARPA) High Productivity Computing System (HPCS) program is developing systems that deliver increased value to users at a rate commensurate with the rate of improvement in the underlying technologies. For example, if the relevant technology was silicon, the goal of such a system would be to double in productivity (or value) every 18 months, following Moore's law. The key questions are how we define and measure productivity, and what the underlying technologies that affect productivity are. The goal of this paper is to synthesize from several different productivity models a single model that captures the main features of all the models. In addition we will start the process of putting the model on an empirical foundation by incorporating selected results from the software engineering and high performance computing (HPC) communities. An asymptotic analysis of the model is conducted to check that it makes sense in certain special cases. The model is extrapolated to a HPC context and several examples are explored, including HPC centers, HPC users, and interactive grid computing. Finally, the model hints at a profoundly different way of viewing HPC systems, where the user must be included in the equation, and innovative hardware is a key aspect to lowering the very high costs of HPC software.