Search

Emerging hardware tailored for artificial intelligence (AI) and machine learning (ML) methods provide novel means to couple them with traditional high performance computing (HPC) workflows involving molecular dynamics (MD) simulations. We propose Stream-AI-MD, a novel instance of applying deep learning methods to drive adaptive MD simulation campaigns in a streaming manner. We leverage the ability to run ensemble MD simulations on GPU clusters, while the data from atomistic MD simulations are streamed continuously to AI/ML approaches to guide the conformational search in a biophysically meaningful manner on a wafer-scale AI accelerator. We demonstrate the efficacy of Stream-AI-MD simulations for two scientific use-cases: (1) folding a small prototypical protein, namely ββα-fold (BBA) FSD-EY and (2) understanding protein-protein interaction (PPI) within the SARS-CoV-2 proteome between two proteins, nsp16 and nsp10. We show that Stream-AI-MD simulations can improve time-to-solution by ~50X for BBA protein folding. Further, we also discuss performance trade-offs involved in implementing AI-coupled HPC workflows on heterogeneous computing architectures.

Large format displays are commonplace for viewing large scientific datasets. These displays often find their way into collaborative spaces, allowing for multiple individuals to be collocated with the display, though multi-modal interaction with the displayed content remains a challenge. We have begun development of a tablet-based interaction mode for use with large format displays to augment these workspaces.

We propose GROPHECY, a GPU performance projection framework that can estimate the performance benefit of GPU acceleration without actual GPU programming or hardware. Users need only to skeletonize pieces of CPU code that are targets for GPU acceleration. Code skeletons are automatically transformed in various ways to mimic tuned GPU codes with characteristics resembling real implementations. The synthesized characteristics are used by an existing analytical model to project GPU performance. The cost and benefit of GPU development can then be estimated according to the transformed code skeleton that yields the best projected performance. With GROPHECY, users can leap toward GPU acceleration only when the cost-benefit makes sense. The framework is validated using kernel benchmarks and data-parallel codes in legacy scientific applications. The measured performance of manually tuned codes deviates from the projected performance by 17% in geometric mean.

Science gateways have dramatically simplified the work required by science communities to run their codes on TeraGrid resources. Gateway development typically spans the duration of a particular grant, with the first production runs occurring some months after the award and concluding near the end of the project. Scientists use gateways as a means to interface with large resources. Our gateway infrastructure facilitates this by hiding away the various details of the underlying resources and presents an intuitive way to interact with the resource. In this paper, we present our work on GPSI, a general-purpose science gateway infrastructure that can be easily customized to meet the needs of an application. This reduces the time to deployment and improves scientific productivity. Our contribution in this paper is two-fold: to elaborate our vision for a user-driven gateway infrastructure that includes components required by multiple science domains, thus aiding the speedy development of gateways, and presenting our experience in moving from our initial portal implementations to the current effort based on Python [15] and Django [16].

A Science Gateway is a computational web portal that enables scientists to run scientific simulations, data analysis, and visualization through their web browsers. The major problem of building a science gateway on TeraGrid is how to deploy scientific applications rapidly on computational resources and expose these applications as web services to scientists. In this paper we propose a novel science application framework that can greatly accelerate the development cycle of science gateway systems. This framework enables science gateway developers to import their domain-specific scientific workflow scripts and generate Web 2.0 gadgets for running these application workflows and visualizing the output from workflow executions without writing any web related code. By assembling these application-specific gadgets and some common gadgets predefined in the framework for workflow management, developers can easily set up a customized computational science gateway to meet community requirements. We demonstrate the utility of the framework with an example from computational biochemistry.

The Social Informatics Data Grid (SIDGrid) is a new cyberinfrastructure designed to transform how social and behavioral scientists collect and annotate data, collaborate and share data, and analyze and mine large data repositories. The major design goals for the SIDGrid are to integrate those commonly used social and behavior science tools and provide researchers an easy-to-use web interface to run these data intensive applications efficiently on TeraGrid resources. SIDGrid is also a collaborative environment where scientists can share experimental data and analysis results with their team members.OpenSocial, a social networking framework initiated by Google, provides a Web 2.0 approach to integration of web applications and building collaborative cyber environments. Using OpenSocial, we present a new application framework for SIDGrid that enables scientists to define their analysis tools in XML and generate application gadgets as a Web 2.0 interface for running analytical workflows on TeraGrid. Based on this framework, we have developed a new SIDGrid science gateway to improve the user's experience and simplify SIDGrid application management and development of collaborative web applications.

Increasingly massive datasets produced by simulations beg the question How will we connect this data to the computational and display resources that support visualization and analysis? This question is driving research into new approaches to allocating computational, storage, and network resources. In this paper we explore potential solutions that couple system resources in new ways.Examples of what we mean by resource-coupled computations abound. For example, remote visualization is an activity that may couple data and large computation resources at the shared facility to client software and display hardware at the remote site. In situ analysis and visualization contemporaneously merges simulation and analysis onto the shared resource of the supercomputing platform. Co-analysis approaches seek to directly couple simulations running on a primary supercomputer to live analysis running on an optimized visualization and analysis platform over a high-performance network.Consequently, we are working on a systems approach to modeling the end-to-end activity of extracting understanding from computational models. In this paper we present our methods and results from experiments.