Exaflops Performance Research Articles

ЭВОЛЮЦИЯ ЦИФРОВОЙ НЕФТЕГАЗОВОЙ ЭКОСИСТЕМЫ ОТ СУПЕРКОМПЬЮТИНГА К МЕТАКОМПЬЮТИНГУ

The subject of the study is the consideration of trends in the evolution of the digital oil and gas ecosystem from supercomputing to metacomputing. The key factor constraining the innovative development of the Russian oil and gas complex is the lack of high-performance computing systems. The key oil and gas tasks requiring the involvement of super- and metacomputing of the peta and exaflop performance levels (1015 and 1018 floatingpoint operations per second, respectively) include the tasks of managing the "digital twin" of the oil and gas complex as a whole, the world's largest West Siberian oil and gas production center and unique oil and gas real-time monitoring of oil and gas fields, and training of highly qualified oil and gas specialists in this field.

Solving the Sampling Problem of the Sycamore Quantum Circuits.

We study the problem of generating independent samples from the output distribution of Google's Sycamore quantum circuits with a target fidelity, which is believed to be beyond the reach of classical supercomputers and has been used to demonstrate quantum supremacy. We propose a method to classically solve this problem by contracting the corresponding tensor network just once, and is massively more efficient than existing methods in generating a large number of uncorrelated samples with a target fidelity. For the Sycamore quantum supremacy circuit with 53 qubits and 20 cycles, we have generated 1×10^{6} uncorrelated bitstrings s which are sampled from a distribution P[over ^](s)=|ψ[over ^](s)|^{2}, where the approximate state ψ[over ^] has fidelity F≈0.0037. The whole computation has cost about 15h on a computational cluster with 512 GPUs. The obtained 1×10^{6} samples, the contraction code and contraction order are made public. If our algorithm could be implemented with high efficiency on a modern supercomputer with ExaFLOPS performance, we estimate that ideally, the simulation would cost a few dozens of seconds, which is faster than Google's quantum hardware.

Parallel Programming

As a source for the introduction of parallel programming models, the article substantiates the development of the technology of VLSI and the corresponding change in the architecture of computing systems in the direction of increasing the parallelism of processing. That is, if the architecture changes significantly, then the parallel programming model should change in order to reduce the difficulty of effectively mapping program to hardware resources. A model of parallel programming on shared memory and synchronization based on FE-bits of shared memory words is considered. The architecture of the computing system of the exaflops performance level is also proposed, for which the considered programming model is adequate. The article tries to convey to the reader that this architecture originated as a synthesis of the development of VLSI technology, architecture of parallel systems and high-production parallel programming. It is indicated that when implementing the proposed architecture and programming model, an architecture with data flow processing is implemented. A parallel programming model for computing systems with local connections and optimal graphs of machine-to-machine connections are presented.

M2C: A Massive Performance and Energy Throttling Framework for High-Performance Computing Systems

At the Petascale level of performance, High-Performance Computing (HPC) systems require significant use of supercomputers with the extensive parallel programming approaches to solve the complicated computational tasks. The Exascale level of performance having 1018 calculations per second is another remarkable achievement in computing with a fathomless influence on everyday life. The current technologies are facing various challenges while achieving ExaFlop performance through energy-efficient systems. Massive parallelism and power consumption are vital challenges for achieving ExaFlop performance. In this paper, we have introduced a novel parallel programming model that provides massive performance under power consumption limitations by parallelizing data on the heterogeneous system to provide coarse grain and fine-grain parallelism. The proposed dual-hierarchical architecture is a hybrid of MVAPICH2 and CUDA, called the M2C model, for heterogeneous systems that utilize both CPU and GPU devices for providing massive parallelism. To validate the objectives of the current study, the proposed model has been implemented using bench-marking applications including linear Dense Matrix Multiplication. Furthermore, we conducted a comparative analysis of the proposed model by existing state-of-the-art models and libraries such as MOC, kBLAS, and cuBLAS. The suggested model outperforms existing models while achieving massive performance in HPC clusters and can be considered for emerging Exascale computing systems.

Программная поддержка модели BSF

The paper is devoted to the software support of the parallel computation model BSF (Bulk Synchronous Farm) intended to iterative algorithms with high computational complexity, which developed for multiprocessor systems with distributed memory of exaflops performance level. Software support includes parallel BSF-skeleton and BSF-Studio web-application. The definition and classification of parallel skeletons are given. The logic and file structure of new BSF-skeleton are described. The BSF-skeleton is a collection of C++ source code files used for BSF-programs development. A detailed description of the design and implementation of the BSF-Studio web-application, which is a visual constructor of BSF-programs, is given. BSF-Studio provides compilation and execution of program code.

Generalized Parallel Computational Schemes for Time-Consuming Global Optimization

This paper addresses computationally intensive global optimization problems, for solving of which the supercomputing systems with exaflops performance can be required. To overcome such computational complexity, the paper proposes the generalized parallel computational schemes, which may involve numerous efficient parallel algorithms of global optimization. The proposed schemes include various ways of multilevel decomposition of parallel computations to guarantee the computational efficiency of supercomputing systems with shared and distributed memory multiprocessors with thousands of processors to meet global optimization challenges.

A data‐driven approach to modeling power consumption for a hybrid supercomputer

SummaryPower consumption of current High Performance Computing systems has to be reduced by at least one order of magnitude before they can be scaled up towards ExaFLOP performance. While we can expect novel hardware technologies and architectures to contribute towards this goal, significant advances have to come also from software technologies such as proactive and power‐aware scheduling, resource allocation, and fault‐tolerant computing. Development of these software technologies in turn relies heavily on our ability to model and accurately predict power consumption in large computing systems. In this paper, we present a data‐driven model of power consumption for a hybrid supercomputer (which held the top spot in the Green500 ranking in June 2013) that combines CPU, GPU, and MIC technologies to achieve high levels of energy efficiency. Our model takes as input workload characteristics—the number and location of resources that are used by each job at a certain time—and calculates a predicted power consumption at the system level. The model is application‐code‐agnostic and is based solely on a data‐driven predictive approach, where log data describing the past jobs in the system are employed to estimate future power consumption. For this, three different model components are developed and integrated. The first employs support vector regression to predict power usage for jobs before these are started. The second uses a simple heuristic to predict the length of jobs, again before they start. The two predictions are then combined to estimate power consumption due to the job at all computational elements in the system. The third component is a linear model that takes as input the power consumption at the computing units and predicts system‐wide power consumption. Our method achieves highly‐accurate predictions starting solely from workload information and user histories. The model can be applied to power‐aware scheduling and power capping: alternative workload dispatching configurations can be evaluated from a power perspective and more efficient ones can be selected. The methodology outlined here can be easily adapted to other HPC systems where the same types of log data are available.

On the role and challenges of CFD in the aerospace industry

ABSTRACTThis article examines the increasingly crucial role played by Computational Fluid Dynamics (CFD) in the analysis, design, certification, and support of aerospace products. The status of CFD is described, and we identify opportunities for CFD to have a more substantial impact. The challenges facing CFD are also discussed, primarily in terms of numerical solution, computing power, and physical modelling. We believe the community must find a balance between enthusiasm and rigor. Besides becoming faster and more affordable by exploiting higher computing power, CFD needs to become more reliable, more reproducible across users, and better understood and integrated with other disciplines and engineering processes. Uncertainty quantification is universally considered as a major goal, but will be slow to take hold. The prospects are good for steady problems with Reynolds-Averaged Navier-Stokes (RANS) turbulence modelling to be solved accurately and without user intervention within a decade – even for very complex geometries, provided technologies, such as solution adaptation are matured for large three-dimensional problems. On the other hand, current projections for supercomputers show a future rate of growth only half of the rate enjoyed from the 1990s to 2013; true exaflop performance is not close. This will delay pure Large-Eddy Simulation (LES) for aerospace applications with their high Reynolds numbers, but hybrid RANS-LES approaches have great potential. Our expectations for a breakthrough in turbulence, whether within traditional modelling or LES, are low and as a result off-design flow physics including separation will continue to pose a substantial challenge, as will laminar-turbulent transition. We also advocate for much improved user interfaces, providing instant access to rich numerical and physical information as well as warnings over solution quality, and thus naturally training the user.

Online Monitoring Systems for Performance Fault Detection

To achieve the exaFLOPS performance within a contained power budget, next generation supercomputers will feature hundreds of millions of components operating at low- and near-threshold voltage. As the probability that at least one of these components fails during the execution of an application approaches certainty, it seems unrealistic to expect that any run of a scientific application will not experience some performance faults. We believe that there is need of a new generation of light-weight performance and debugging tools that can be used online even during production runs of parallel applications and that can identify performance anomalies during the application execution. In this work we propose the design and implementation of a monitoring system that continuously inspects the evolution of running applications and the health of the system. To achieve minimum runtime overhead while maintaining the desired level of flexibility, we propose a decoupled approach in which accurate monitoring is performed at kernel-level while performance anomaly disambiguation and corrective actions are performed at user-level. We evaluate our monitoring system on a 128-core AMD Interlagos system: First, we show that the runtime overhead of the monitoring system is negligible (0-2%). Then we show how our system can be used to precisely identify performance faults in three different scenarios: OS noise, application co-scheduling and dynamic power capping.