Helgi Adalsteinsson

A Simulator for Large-Scale Parallel Computer Architectures

Efficient design of hardware and software for large-scale parallel execution requires detailed understanding of the interactions between the application, computer, and network. The authors have developed a macro-scale simulator SST/macro that permits the coarse-grained study of distributed-memory applications. In the presented work, applications using the Message Passing Interface MPI are simulated; however, the simulator is designed to allow inclusion of other programming models. The simulator is driven from either a trace file or a skeleton application. Trace files can be either a standard format Open Trace Format or a more detailed custom format DUMPI. The simulator architecture is modular, allowing it to easily be extended with additional network models, trace file formats, and more detailed processor models. This paper describes the design of the simulator, provides performance results, and presents studies showing how application performance is affected by machine characteristics.

Uncertainty Quantification in MD Simulations. Part II: Bayesian Inference of Force-Field Parameters

This paper explores the inference of small-scale, atomistic parameters, based on the specification of large, or macroscale, observables. Specifically, we focus on estimating a set of force-field parameters for the four-site, TIP4P, water model, based on a synthetic problem involving isothermal, isobaric molecular dynamics (MD) simulations of water at ambient conditions. We exploit the polynomial chaos (PC) expansions developed in Part I as surrogate representations of three macroscale observables, namely density, self-diffusion, and enthalpy, as a function of the force-field parameters. We analyze and discuss the use of two different PC representations in a Bayesian framework for the inference of atomistic parameters, based on synthetic observations of three macroscale observables. The first surrogate is a deterministic PC representation, constructed in Part I using nonintrusive spectral projection (NISP). An alternative strategy exploits a nondeterministic PC representation obtained using Bayesian infere...

Uncertainty Quantification in MD Simulations. Part I: Forward Propagation

This work focuses on quantifying the effect of intrinsic (thermal) noise and parametric uncertainty in molecular dynamics (MD) simulations. We consider isothermal, isobaric MD simulations of TIP4P (or four-site) water at ambient conditions,

Using simulation to design extremescale applications and architectures: programming model exploration

T=298

Design patterns for multiphysics modeling in Fortran 2003 and C

K and

A Stochastic Multiscale Coupling Scheme to Account for Sampling Noise in Atomistic-to-Continuum Simulations

P=1

Components for atomistic-to-continuum multiscale modeling of flow in micro- and nanofluidic systems

atm. Parametric uncertainty is assumed to originate from three force-field parameters that are parametrized in terms of standard uniform random variables. The thermal fluctuations inherent in MD simulations combine with parametric uncertainty to yield nondeterministic, noisy MD predictions of bulk water properties. Relying on polynomial chaos (PC) expansions, we develop a framework that enables us to isolate the impact of parametric uncertainty on the MD predictions and control the effect of the intrinsic noise. We construct the PC representations of quantities of interest (QoIs) using two different approaches: nonintrusive spectral projection (NISP) and Bayesian inference. We verify a priori the legitimacy of the NISP approach by verifying that the...

Computational and experimental study of nanoporous membranes for water desalination and decontamination.

A key problem facing application developers is that they are expected to utilize extreme levels of parallelism soon after delivery of future leadership class machines, but developing applications capable of exposing sufficient concurrency is a time consuming process requiring experimentation. At the same time, due to the expense of building and operating an exascale machine, it will be necessary to apply tighter engineering margins to their design. Simple metrics such as the computation-communication ratio will not sufficiently specify machine requirements. Simulation fills this gap, allowing the study of extreme-scale architectures with the explicit inclusion of the complex interactions between the various hardware and software components, and can be used for correctness-checking as well as performance estimation. The simulator we discuss in this paper can be driven by reading trace files, typically generated by an actual application that has been run on real hardware, or by using a skeleton application. The skeleton application is designed to have the control flow of a real application, but with expensive computations and large data transfers replaced by discrete events for which the timings are determined by models. Using skeleton applications, we can predict application performance at levels of parallelism unobtainable on any current computational platform. The skeleton application can be modified to experiment with different communication strategies and programming models. Since the machine being simulated is in our control, we can experiment with different network topologies, routing algorithms, bandwidths, latencies, failure modes, core-to-node ratios, etc. In this paper, we use the Structural Simulation Toolkit macroscale components for coarse-grained simulation to illustrate the exploration of alternative programming models at extreme scale.

Data-free inference of the joint distribution of uncertain model parameters

We present three new object-oriented software design patterns in Fortran 2003 and C++. These patterns integrate coupled differential equations, facilitating the flexible swapping of physical and numerical software abstractions at compile-time and runtime. The Semi-Discrete pattern supports the time advancement of a dynamical system encapsulated in a single abstract data type (ADT). The Puppeteer pattern combines ADTs into a multiphysics package, mediates interabstraction communications, and enables implicit marching even when nonlinear terms couple separate ADTs with private data. The Surrogate pattern emulates C++ forward references in Fortran 2003. After code demonstrations using the Lorenz equations, we provide architectural descriptions of our use of the new patterns in extending the Rouson et al. [2008a] Navier-Stokes solver to simulate multiphysics phenomena. We also describe the relationships between the new patterns and two previously developed architectural elements: the Strategy pattern of Gamma et al. [1995] and the template emulation technique of Akin [2003]. This report demonstrates how these patterns manage complexity by providing logical separation between individual physics models and the control logic that bridges between them. Additionally, it shows how language features such as operator overloading and automated memory management enable a clear mathematical notation for model bridging and system evolution.

Evaluating Performance Optimizations of Large-Scale Genomic Sequence Search Applications Using SST/macro.

We present a methodology to assess the predictive fidelity of multiscale simulations by incorporating uncertainty in the information exchanged between the atomistic and continuum simulation components. Focusing on uncertainty due to finite sampling in molecular dynamics (MD) simulations, we present an iterative stochastic coupling algorithm that relies on Bayesian inference to build polynomial chaos expansions for the variables exchanged across the atomistic-continuum interface. We consider a simple Couette flow model where velocities are exchanged between the atomistic and continuum components. To alleviate the burden of running expensive MD simulations at every iteration, a surrogate model is constructed from which samples can be efficiently drawn as data for the Bayesian inference. Results show convergence of the coupling algorithm at a reasonable number of iterations. The uncertainty associated with the exchanged variables significantly depends on the amount of data sampled from the MD simulations and...