Padma Raghavan - US grants

9502594 Raghavan The primary objective of Raghavan's research is to develop algorithms and software for a scalable, fully parallel solution of large, sparse linear systems of equations using direct methods; the solver developed will be used in a key application both to demonstrate the significance of the work and to help study and resolve issues of interface. A secondary objective is to provide tools suitable for other areas of scientific computing obtained as variants of some of the algorithms developed.

CDA-9529459 Michael W. Berry Jens Gregor Mark T. Jones James S. Plank Padma Raghavan University of Tennessee An ATM networked 12 high-performance workstations supports several research projects in the Computer Science Department at the University of Tennessee, including in particular: Scientific Applications in a Distributed Computing Environment, PET Image Reconstruction in a Distributed Environment, Parallel Algorithms and Software for Unstructured Mesh Computations, Fast Checkpointing in Parallel Environments, and Parallel Sparse Matrix Computations. This computing laboratory provides required resources of CPU time, local memory, communication bandwidth and disk storage for the computationally intensive projects. Some of the research tasks require this dedicated laboratory for their exceptionally long runtimes. The systems project requires access to such an environment to run meaningful experiments. In addition, the dedicated laboratory facilities may be reconfigured or subjected to simulated faults without concern about the effects on other users. The research projects will produce public-domain software. Because several government and industrial sites will soon have facilities similar to that which we propose, regular access to such a laboratory is essential for quality software research and development.

9818334
Dongarra, Jack J.
Plank, James S.
University of Tennessee

Enabling Technology for High-Performance Heterogeneous Clusters Focusing on Grid Middleware, Fault Tolerance and Sparse Matrix Computations

This research instrumentation enables research projects in:

- Harnessing Cluster Resources for Distributed Scientific Computing,
- Fast and Portable Checkpointing in Clusters and Clusters-of-Clusters, and
- Tools for Large-Scale Sparse Matrix Applications on Clusters.

To support the aforementioned projects, this award contributes to the purchase of a 32-node high performance cluster, some switches, workstations, and interface to existing clusters and visualization lab at
the University of Tennessee. High-performance, low-latency clusters assembled from commodity computers and interconnects are clearly the cost-effective alternative to parallel supercomputers. But the software environment on such clusters is primitive at best; there is an obvious need of tools for application development and cluster management. The three projects in this proposal address this need. Their common goal is the development of enabling technology for advanced scientific computing applications on large-scale clusters and heterogeneous clusters-of-clusters. The proposed instrumentation is for a high-performance networked cluster of workstations. This cluster will be connected to two small clusters (available in the department) to provide a sizable, heterogeneous ``cluster-of-clusters'' with visualization capabilities. This cluster-of-clusters parallel platform will be used for algorithm, software and tool development research in the three projects.

ABSTRACT
Proposal: 0075792
PI: Michael Langston


An adaptive computing system (ACS) offers a revolutionary combination of the performance of custom hardware and the flexibility of software by employing reconfigurable technology. A key feature of an ACS is the reconfigurable processing element, which, in the current generation, is a Field-Programmable Gate Array (FPGA) chip. This research project investigates the impact of an ACS in the context of a high-performance computational grid with clusters-of-workstations, shared memory multi-processors and rapid interconnects. Suites of fast estimators are devised using approximation algorithms for FPGA mapping and partitioning. An assortment of algorithmic methods is applied. A major focus is on new heuristic and optimization strategies designed to exploit emergent mathematical techniques. Supporting software tools are also developed, with an emphasis placed on portability. Implementation testbeds are built around edge-based segmentation and related problems common to a variety of image processing applications.

Robust Limited Memory Hybrid Sparse Solvers

Sparse linear solvers can be broadly classified as being either 'direct' or 'iterative.' Direct solvers are based
on a factorization of the associated sparse matrix and are extremely robust. However, their memory requirements grow as a non-linear function of the matrix dimension because original zeroes fill-in during factorization.
The Krylov subspace (KSP) family of iterative methods are memory scalable, but their convergence
can be slow or fail altogether. This project concerns developing scalable hybrids than can be parameterized
to model the range from pure iterative to pure direct methods. We propose to develop parallel algorithms
and software engineering methods aimed at providing robust, limited memory hybrid solvers that satisfy the
computational demands of a variety of applications.
On the algorithmic front, our focus is on hybrids obtained by preconditioning KSP solvers using suitable
incomplete matrix factors. Such preconditioners are robust and widely applicable, but until recently they
were considered unsuitable for parallel computing. The main reason is that the sparse triangular solves for
applying the preconditioner become a bottleneck due to the relatively high latency of communication. We
have recently developed a latency tolerant 'selective-inversion' scheme that overcomes this problem to yield
an efficient and scalable implementation. In this project, we propose developing parallel sparse factorization
techniques that are efficient for the entire spectrum of fill-in. We will develop a new 'supernodal diagonal
row block' formulation for scalable incomplete factorization. We will also consider innovative ways of
combining symbolic (level of fill) and numeric (threshold) strategies to specify fill-in to be either retained
or discarded. Additionally, our algorithmic framework enables us to provide a single, unified, extensible
implementation of hybrids for symmetric positive definite, symmetric indefinite, and nonsymmetric systems.
On the software front, we define a new 'usage model' based 'reverse engineering' process to develop a high-performance domain specific solver as a smart composite of several methods. Our premise is that the right composite solver is domain specific; substantial performance gains can be realized by selecting the right combination of underlying methods to match linear system attributes. We will obtain a uniform interface to a variety of parallel sparse solver software by developing an object-oriented sparse template library that utilizes parameterized polymorphism. Composites will be instantiated by using this template library and a scripting language that supports parallel computing using MPI.
Our design goals and performance targets will be keyed to three large-scale computational science applications. The first concerns computational methods for advanced optimization; this application requires robust indefinite solvers. The second is a structural mechanics application for modeling cracks and fractures. The third application involves large sparse eigenvalue problems that arise in quantum molecular dynamics.
Our project represents a concerted effort to resolve critical research issues in the area of parallel sparse
matrix computations. Our goal is to develop the next generation of sparse solvers by combining research in
parallel algorithms and software engineering.

This proposal was received in response to NSE, NSF-0019. A novel approach for the fabrication of atomistic electronic devices is proposed which, if realized, will have important industrial applications. The theoretical methods involve simulations using quantum tight-binding molecular dynamics scheme that can be used to accurately treat interactions in carbon systems at the nanoscale level. Large scale simulations will be performed using novel parallel computer algorithms using a synergistic interdisciplinary collaboration. Simulation results can be used as a guide in the experimental investigations.

Although the present electronic technology is dominated by silicon, it is becoming clear that Si based electronic devices cannot be relied on to sustain the current pace of miniaturization. It is
becoming clear that a new class of molecularly perfect materials are needed to make these new devices. Single-wall carbon nanotubes are one such material that are expected to execute a ``quantum'' leap i the area of nanoscale electronics, computers, and materials. A focussed effort to lay the foundation for fullerene and nanotube based molecular electronics which will revolutionize the electronics and computer industries is proposed. The emphasis is on the modeling and simulations which will be used to guide experimental efforts to realize these devices. The theoretical method for the treatment of these systems contains many state-of-the-art features, making it ideally suited for studying these systems.

EIA 02-02007
Das, Chita R.
Acharya, Raj; Giles, C. L.; Irwin, Mary Jane; Plassman, Paul E.
Pennsylvania State University
(I^3\)C: An Infrastructure for Innovation in Information Computing

This proposal, advancing the state-of-the-art in cluster computing, addresses a broad spectrum of interactive and archival information processing issues that require intensive computational capabilities. A high performance compute-engine (clusters), the Access Net, Storage Network (file and database server), and a Media Lab (multimedia and graphics equipment, workstations, laptops, PDAs, and software) comprise the infrastructure that will support the following three core areas, with respective research activities:
1. Applications
a. Computational Science
b. Digital Immortality
c. Bioinformatics
2. System Software
a. Cluster Scheduling
b. Shared I/O Support
c. Fault Tolerance
3. Architecture
a. QoS Support for Multimedia and E-Commerce Applications in Clustered-Servers
b. Design of Energy-Efficient Hardware and Software Optimizations
c. Energy Perspective Optimization of Different Layers of the Wireless Protocol Stack
The project focuses on three application areas (Computational Science, Digital Immortality, and Bioinformatics) to examine common research issues, which in turn drive the Software and Architecture research for providing holistic solutions to challenging problems. A research team, consisting of 25 investigators from three different colleges, will benefit from this infrastructure by strengthening their interdisciplinary research through cross-fertilization of new ideas.

This award is made under the Information Technology Research initiative and is funded jointly by the Division of Materials Research and the Advanced Computational Infrastructure Research Division.

This collaborative research project involves two materials scientists, a computer
scientist, a mathematician, and two physicists from academia, industry and a national laboratory. The project is a synergistic effort that leverages the overlapping and complimentary expertise of the researchers in the areas of scalable parallel scientific computing, first-principles and atomistic calculations, computational thermodynamics, mesoscale microstructure evolution, and macroscopic mechanical property modeling. The main objective of the proposal is to develop a set of integrated computational tools to predict the relationships among the chemical, microstructural, and mechanical properties of multicomponent materials using technologically important aluminum-based alloys as model materials. A prototype GRID-enabled software will be developed for multicomponent materials design with efficient information exchange between design stages. Each design stage will incorporate effective algorithms and parallel computing schemes. Four computational components will be integrated, these are: (1) first-principles calculations to determine thermodynamic properties, lattice parameters, and kinetic data of unary, binary and ternary compounds; (2) CALPHAD data optimization computation to extract thermodynamic properties, lattice parameters, and kinetic data of multicomponent systems combining results from first-principles calculations and experimental data; (3) multicomponent phase-field modeling to produce microstructure; and (4) finite element analysis to obtain the mechanical response from the simulated microstructure. The research involves a parallel effort in information technology with two main components: (1) advanced discretization and parallel algorithms, and (2) a software architecture for distributed computing system. The first component includes: (a) a coupling of spectral and finite element approximations, (b) local adaptivity and multi-scale resolution, (c) high order stable semi-implicit in time schemes, (d) parallelization through domain decomposition, and (e) scalable sparse system solvers. The second component involves computational GRID-enabled software for the overall design process; this software architecture enables the use of geographically distributed high performance parallel computing resources to reduce application turnaround time while providing a flexible client-server interface that allows multiple design cycles to proceed.

The research project will be integrated with education and training of graduate students in the broad area of computational science and engineering through the participation of students and the PIs in the "High Performance Computing Graduate Minor" offered through the Institute of High Performance Computing at The Pennsylvania State University. Existing programs at Penn State will be used to integrate undergraduates into the project.
%%%
This award is made under the Information Technology Research initiative and is funded jointly by the Division of Materials Research and the Advanced Computational Infrastructure Research Division.

This collaborative research project involves two materials scientists, a computer
scientist, a mathematician, and two physicists from academia, industry and a national laboratory. The project is a synergistic effort that leverages the overlapping and complimentary expertise of the researchers in the areas of scalable parallel scientific computing, first-principles and atomistic calculations, computational thermodynamics, mesoscale microstructure evolution, and macroscopic mechanical property modeling. The main objective of the proposal is to develop a set of integrated computational tools to predict the relationships among the chemical, microstructural and mechanical properties of multicomponent materials using technologically important aluminum-based alloys as model materials. Prototype GRID-enabled software will be developed for multicomponent materials design. Effective algorithms and parallel computing schemes will be incorporated into the design. The GRID-enabled software allows geographically distributed high performance parallel computing resources to be harnessed bringing greater computational power to bear on a given problem and enabling practical application of these computational tools. The prototype software, with improved predictive power in multicomponent materials design, may enable scientists to develop new materials with unique properties and to tailor existing materials for better performance.

The research project will be integrated with education and training of graduate students in the broad area of computational science and engineering through the participation of students and the PIs in the "High Performance Computing Graduate Minor" offered through the Institute of High Performance Computing at The Pennsylvania State University. Existing programs at Penn State will be used to integrate undergraduates into the project.
***

EIA-0221916
Menon, Madhu
University of Kentucky

TITLE: ITR Large Scale Quantum Mechanical Simulations of Nonomechanics


Large scale quantum mechanical simulations of nanomechanics of carbon
nanotubes involving atoms in excess of 10,000 is underway using a
collaborative effort involving material and computer
scientists. Parallel software tools are being developed to accomplish
this. The tools are based on a novel pipelined, parallel architecture
designed to harness grid computing. The goal of the simulations is to
study the potential benefits of nanomechanical applications of carbon
nanotubes and use the results to guide experimental investigations.

The Society for Industrial and Applied Mathematics(SIAM) will present the Eleventh Conference on Parallel Processing for Scientific Computing. This series of conferences has played a key role in promoting parallel scientific computing and parallel numerical algorithms. The conference is distinguished by its emphasis on the information technology aspects of scientific computing on parallel machines and provides a forum for communication among the scientific computing, information technology, and computational science and engineering communities.

The bulk of this grant will support small travel grants to student and junior researchers. The grant will also support travel grants and small monetary awards to winners in a student paper competition and a poster competition. All award decisions will be made by the conference organizing committee (the PI is a Co-chair). SIAM will advertise and coordinate these activities and if this proposal is funded, NSF support will be clearly acknowledged at the conference and its web-site and related publications.

This proposal seeks to address these two primary challenges by developing adaptive software tools to
co-manage quality-performance-power tradeoffs. First, we will develop combinatorial and statistical adaptive techniques to select methods dynamically, delivering the improved performance while producing a solution that meets application quality requirements. Next, using an annotated model of computation and communication costs and sparse data access patterns, we will develop techniques for power reduction without
performance impairment. For example, power savings can be significant even when relatively minor load
imbalances among processors are exploited. These imbalances can easily be on the order of trillions of
CPU cycles, and power consumption can be tuned through dynamic voltage scaling (DVS, where both the
clock frequency and the supply voltage are tuned) for lightly/heavily loaded processors. More importantly,
resulting insights can lead to future systems where the power budget is directed effectively over processor-memory interconnect subsystems to improve application performance. We plan to implement our techniques by developing an adaptive component software system on high-end multiprocessors

The Industry/University Cooperative Research Center for Computational Materials Design joins Penn State and Georgia Tech to substantially impact progress towards systems-based materials design by promoting research programs of interest to both industry and universities, to enhance the infrastructure of computational materials research in the nation, to explore and extend the interface between engineering systems design, information technology and physics-based simulation of process-structure and structure-property relations of materials, to improve the intellectual capacity of the workforce through industrial participation and conduct of high quality research projects, and to develop curriculum in computational and systems design aspects of materials.

This project seeks to enable model-driven optimizations spanning
multiple levels of a computing system including the architecture,
compiler, algorithm and application layers, for multiple objectives
such as performance, power and productivity. A primary goal is to
develop a comprehensive framework for model-driven multilevel,
multiobjective optimizations with a focus on chip multiprocessors
(CMPs) and large-scale, sparse engineering and scientific applications.
Key activities concern developing (i) parameterized models to
compose models of the application, architecture and compiler
transformations, (ii) an optimization framework to determine
multiobjective, optimal or pareto-optimal designs while modeling
uncertainties, and (iii) undergraduate and graduate courses on the
methodology for multilevel optimizations of computing systems,

The proposed techniques yield metrics at coarse- and medium-scales
that can be used with stochastic optimization techniques to determine
optimal design choices. The medium-scale metrics are obtained by
simulating a concatenated discrete time Markov Chain model (C-DT-MCM)
which incorporates both the deterministic and stochastic aspects
of multilevel optimizations and their impacts. Such C-DT-MCMs can
be simulated very efficiently to obtain traces which can then be
compared using statistical techniques with those from detailed
hardware simulation. Using this approach, only promising design
options need be studied in detail, using current modalities, such
as detailed hardware simulators, which can be prohibitively slow
for larger CMP architectures.

Proposal #: CNS 08-21527
PI(s): Raghavan, Padma
Chen, Long-Quing; Hudson, Peter J.; Kandemir, Mahmut T.; Smith, Brian K.
Institution: Pennsylvania State University
University Park, PA 16802-700
Title: MRI/Acq.: Acq.of A Scalable Instrument for Discovery through Computing
MRI Acquisition of a Scalable Instrument for Discovery through Computing

This award from the Major Research Instrumentation Program (MRI)
provides funds for the acquisition of a terascale advanced computing
instrument at the Pennsylvania State University. The instrument
will enable researchers from seven disciplines (biological, materials
and social sciences, computer and information science, engineering,
education, and geosciences), to perform virtual experiments toward
discovery and design through computing. Research projects concern:
predictive network modeling of infectious disease dynamics, designing
new piezoelectric materials, designing next-generation chip
multiprocessors, modeling human interactions to promote learning
in virtual communities, and the development of a critical zone
environmental observatory. Despite their diversity, these projects
share computational scalability challenges to be addressed for
enabling scientific advances that often depend on solving large
problems representing a sufficient level of detail and complexity.
The instrument will form the core of a multidisciplinary collaborative
environment to enable transformative approaches to address the
challenges of scaling at multiple levels. It will support a set of
integrated research, education, training, and outreach activities
to: (i) enable collaborative scaling across projects through the
transfer of scaling approaches from one domain into another, while
addressing algorithmic, system, or instrument scaling challenges
within individual projects, (ii) promote technology-transfer through
industrial partnerships, and (iii) grow and enhance the diversity
of the limited computational science talent pool.

Toward a Linear Time Sparse Solver with Locality-Enhanced Scalable Parallelism
Padma Raghavan
Many computational science simulations concern the numeric solution of partial differential equations using implicit or semi-implicit methods. Such simulations often involve nonlinear systems where the dominant costs are those for sparse linear system solution. Sparse solvers present an array of performance challenges on current and future generation multicore chip-multiprocessor architectures, and their networked ensembles into massively parallel processing systems. A new structured hybrid algorithm is sought to enable reliable and scalable solution. An inner-outer tree-structured scheme is formulated by exploiting geometry for tearing and interlacing of the graph of the coefficient matrix. Research issues concern scalable parallelism, hardware latency-masking mapping of data, and geometric mechanisms to convey numeric properties of the coefficient matrix to accelerate the solution process.

Emerging high-end computing platforms, such as leadership-class machines at the petascale, provide new horizons for complex modeling and large-scale simulations. These machines are used to execute data intensive applications of national interest such as climate modeling, cosmic microwave background radiation, and astrophysical thermonuclear flashes. While these systems have unprecedented levels of peak computational power and storage capacity, a critical challenge concerns the design and implementation of scalable I/O (input-output) system software (also called I/O stack) that makes it possible to harness the power of these systems for scientific discovery and engineering design. Unfortunately, currently, there are no available mechanisms that accommodate I/O stack-wide, application-level QoS (quality-of-service)specification, monitoring, and management.

This project investigates a revolutionary approach to the QoS-aware management of the I/O stack using feedback control theory, machine learning, and optimization. The goal is to maximize I/O performance and thus improve overall performance of large scale applications of national interest. The project uses (1) machine learning and optimization to determine the best decomposition of application-level QoS to sub-QoSs targeting individual resources, and (2) feedback control theory to allocate shared resources managed by the I/O stack such that the specified QoSs are satisfied throughout the execution. The project tests the developed I/O stack enhancements using the workloads at NCAR, LBNL and ANL systems. It also involves two efforts in broadening participation: CISE Visit in Engineering Weekends (VIEW) and NASA-Aerospace Education Services Project (NASA-AESP) at the Center for Science and the Schools (CSATS).

Center for Computational Materials Design (CCMD)

IIP-1034965 Pennsylvania State University (PSU)
IIP-1034968 Georgia Tech (GT)

This is a proposal to renew the Center for Computational Materials Design (CCMD), an I/UCRC center that was created in 2005. The lead institution is Pennsylvania State University, and the research partner is Georgia Tech. The main research mission of the CCMD is to develop simulation tools and methods to support materials design decisions and novel methods for collaborative, decision-based systems robust design of materials.

The intellectual merit of CCMD is based on the integration of multiscale, interdisciplinary computational expertise at PSU and GT. CCMD provides leadership in articulating the importance of integrated design of materials and products to industry and the broad profession of materials engineering; and is developing new methods and algorithms for concurrent design of components and materials.

CCMD has operated successfully in Phase I, and has helped develop a partnership amongst academe, industry and national laboratories. Based on feedback received from the various members, CCMD has outlined in the renewal proposal research thrusts and initiatives for Phase II; and has also identified gaps that will be addressed as research opportunities in Phase II.

CCMD will have a large impact on how industry addresses material selection and development. The expanded university/industry interaction of this multi-university center offers all participants a broader view of material design activities in all sectors. CCMD contributes to US competitiveness in computational materials design by educating new generations of students who have valuable perspectives on fundamental modeling and simulation methods, as well as industry-relevant design integration and materials development. CCMD participates in programs at PSU and GT that support K-12 STEM issues, women and underrepresented groups, undergraduate students, and high school teachers. CCMD plans to disseminate research results via papers, conferences and the CCMD website.

The PIs are working on developing and evaluating a data-driven three-phase adaptive, sparse multicore data mining framework for scalable and efficient supervised classification and statistical analysis.

Phase-I seeks to characterize data attributes in terms of sparsity, graph-theoretic structure and geometric and numeric measures toward data transformations with a focus on dimensionality reduction. The goal is to explore the trade-offs between quality of solution (accuracy and precision of classification) and total work (sequential computational costs) toward faster, yet improved methods.

Phase-II operates on the transformed data to increase the degree of fine to coarse grained concurrency while restructuring the data for enhanced reuse and locality of access. This phase provides a weighted annotated graph model of the computations indicating dependencies, data sharing measures and computational costs.

Phase-III utilizes this model to formulate and explore architecture-aware mappings of data mining computations to the multicore processors, including cache and bandwidth aware thread-to-core mappings that consider both performance and power.

The PIs thus seek adaptations to utilize data set attributes, including approximations and concurrency of computations latent in the sparsity structure, toward improved utilization of processor and memory hardware on current and future multicores with larger core counts, complex cache hierarchies and off-chip bandwidth constraints.

Future computer architectures will execute critical applications ranging from climate modeling to bioinformatics to simulations of new materials. Emerging chip multiprocessors are novel architectures that provide high performance and low power consumption. When multiple applications share the same chip multiprocessor, proper management of architectural resources becomes a critical problem. This research targets inter-application management of different types of resources found in a chip multiprocessor using robust and resilient techniques drawn from formal feedback control theory. It enables applications to satisfy their performance requirements and, at the same time, maximizes the utilization of hardware resources.

This research helps to make the transition from conventional architectures to chip multiprocessors smoother, and as a result, maximizes the number of critical applications that can take advantage of these emerging architectures. The transfer of the technology being developed to IBM and HP Labs allows serious testing of the proposed strategies in large-scale architectures built from chip multiprocessors. The results from this project not only present an accurate assessment of feedback control theory in a new domain, but also get integrated into two important software infrastructures, Singularity and Xen, both used extensively by large research communities. This research also accommodates two efforts in broadening participation, CISE Visit in Engineering Weekends (VIEW) and NASA-Aerospace Education Services Project (NASA-AESP) at the Center for Science and the Schools (CSATS).

A large number of "big data" and "big simulation" applications, such as those for determining network models or simulations of partial differential equation models, concern high dimensional data that are sparse. Sparse data structures and algorithms present significant advantages in terms of storage and computational costs. However, with only a few operations per data element, efficient and scalable implementations are difficult to achieve on current and emerging high performance computing systems with very high degrees of core level parallelism, complex node interconnect topology and multicore/manycore nodes with non-uniform memory architectures (NUMA). This proposal develops and evaluates á-embedded graph hardware-software models and attendant data locality-preserving and NUMA-aware application to core/thread mappings to enhance performance and parallel scalability.
Consider an application task graph A, weighted with measures of work and data sharing that is approximately embedded in two or three dimensions, to obtain an á-embedded graph A. Additionally, consider a weighted graph of a HPC system that is naturally assigned coordinates to obtain an á-embedded host graph model H. This proposal develops parallel algorithms to compute interconnect topology-aware mappings of A to H in order to optimize performance measures such as congestion and dilation while preserving load balance. Additionally, at a multicore node in H that is assigned a subgraph of A, (i) sparse data are reordered to enhance parallelism and locality, and (ii) a dynamic fine-grain NUMA-aware task scheduling is applied to respond through work-stealing to core variations in performance from resource conflicts, throttling etc. Finally, through insights gained from á-embedded graph models, sparse matrix algorithms are reformulated to enhance communication avoidance, soft error resilience and data preconditioning. Outcomes include enabling weak scaling to a very large number of cores by extracting parallelism at fine, medium and large-grains, and significantly enhanced fixed and scaled problem efficiencies through locality preservation.
The interconnect topology-aware models and maps hold the potential for impact on very large scale HPC workloads through potential incorporation into the Message Passing Interface for enhanced sparse communications. Additionally, the proposed locality-aware mappings and NUMA-aware scheduling can potentially benefit the very large base of modeling and simulation applications that run on small multicore clusters. Graduate student training is enhanced through a "scale-up" challenge component in an interdisciplinary course on computational science and engineering. High school students are introduced to parallel computing through summer in-residence programs seeking to broaden participation in science and engineering from underrepresented communities.

The proposed research harnesses parallelism to accelerate the
pervasive bioinformatics workflow of detecting genetic variations.
This workflow determines the genetic variants present in an
individual, given DNA sequencing data. The variant detection workflow
is an integral part of current genomic data analysis, and several
studies have linked genetic variants to diseases. Typical instances
of this workflow currently take several hours to multiple days to
complete with state-of-the-art software, and current algorithms and
software are unable to exploit and benefit from even modest levels of
hardware parallelism. Most prior approaches to parallelization and
performance tuning of genomic data analysis pipelines have targeted
computation, I/O, or network data transfer bottlenecks in isolation,
and consequently, are limited in the overall performance improvement
they can achieve. This project targets end-to-end acceleration
methodologies and uses emerging heterogeneous supercomputers to
reduce workflow time-to-completion.

The project focuses on holistic methodologies to accelerate multiple
components within the genetic variant detection workflow. It explores
lightweight data reorganizations at multiple granularities to enhance
locality, investigates compute-, communication-, and I/O task
cotuning, locality-aware load-balancing, and coordinated resource
partitioning to exploit high-performance computing platforms. A key
goal of the proposed research is to design domain-specific
optimizations targeting the massive parallelism and scalability
potential of current heterogeneous supercomputers, so that the
developed techniques can be easily transferred and applied to
dedicated academic cluster and commercial computational environments.
Outreach efforts target undergraduate students through recruiting
workshops and attract them to interdisciplinary graduate programs.
Curriculum development activities emphasize cross-layer parallelism.

For further information, see project web site at
http://sites.psu.edu/XPSGenomics

In computer-aided design and analysis of engineered systems such as automobiles or semiconductor chips, computational models are simulated on high-performance computers to characterize and evaluate key attributes. The sheer scale of such high-performance computing systems, e.g., over 20 billion transistors in Summit (one of the world's fastest supercomputers), increases the likelihood of transient hardware faults from events such as cosmic radiation or processor-chip voltage fluctuations. The likelihood of such errors and their negative impacts are further increased as such simulations are typically long running, and the corruption of a single data field or variable may require weeks to months of re-computations before critical decisions can be made. This project will develop automated approaches that bring fault tolerance to hardware faults for such applications which are widely used not only across multiple industrial sectors but to also increase the predictive power of climate or weather models to aid critical decision making.

Traditional fault-tolerant schemes can be either application-specific, requiring significant programmer effort to redesign or customize large-scale software, or application-agnostic where all or most data are redundantly stored periodically to allow for recovery, thus limiting their scalability due to their significant memory and processing overheads. This project seeks to address these limitations by providing a theoretical foundation for a new class of fault-tolerant schemes that are suitable for the broad array of applications based on iterative numerical simulations that evolve over time on discretized spatial domains. This project is based on the premise that in such physics-based applications, the rate of change of the solution vector components across time steps (iterations) and spatial domains is a key metric to automatically identifying the critical computational variables, monitoring their evolution, and dynamically selecting the type of safeguarding techniques that should be applied. The investigators will pursue three key directions: (i) characterizing the intrinsic resiliency of the application by developing resiliency gradient metrics, (ii) developing and testing fault-tolerance schemes that adapt the level and type of protection to the resiliency gradient with the goal of reducing computational overheads and increasing scalability, and (iii) constructing an automatic online decision-based learning framework for adaptively selecting fault-tolerance methods in relation to the system's ability to use approximate computing and co-scheduling techniques. The investigators will also work closely with application and runtime system developers to seek broader use of this fault tolerance framework, develop specialized undergraduate and graduate curriculum for student training, and offer research experiences to high school students.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

The broader impact/commercial potential of this NSF I-Corps Hubs project is the development of a sustainable, inclusive, innovation ecosystem that will impart shared economic prosperity across the US Mid-South region. The Mid-South Hub’s formative, evidence-driven approach may inform how inclusive innovation ecosystems can be established in regions with nascent economic activity. This knowledge will augment team preparation, maximizing the value extracted from I-Corps training. The Hub will further amplify the downstream successes and sustainability of deep technology ventures, including grant acquisition, fundraising, sustainable revenues, and net-positive liquidity events. Success will be leveraged to influence innovation culture at regional institutions, leading to the incentivization of enterprising faculty, students, and staff. The diverse leadership team and inclusive bench of instructors will increase the Hub’s ability to connect to innovators and mentors of all backgrounds, including those who are disadvantaged or underrepresented. These efforts will include building commercialization capacity at underrepresented institutions. Evidence from the deliberate implementation of a democratized organizational structure will be evaluated for its capability to encourage and preserve the cognitive proximity of ecosystem members. This knowledge will be disseminated via the National Innovation Network to drive translational impacts and economic development, shaping the future of American innovation.<br/><br/>This I-Corps Hubs project is based on the development of a use-inspired incubator that uses data-driven approaches to develop best practices, influence economic policy, inspire adoption of inclusive approaches to innovation, and instruct future programmatic investments. Currently, a gap exists in understanding how regional innovation clusters can unify to drive the development of a prolific innovation ecosystem. Addressing this gap will enable policymakers and government agencies to implement evidence-based approaches to inform programmatic investments and maximize technology commercialization, economic development, and overall national innovation readiness. This consortium of diverse, deep technology-producing institutions from disparate locations within the Mid-South region will leverage the I-Corps program to catalyze technological commercialization, spur economic development, and inform the future of inclusive American innovation. The Hub will prioritize a formative, longitudinal assessment to iteratively optimize key activities, including team recruitment, Regional and National I-Corps training, upstream changes in university innovation culture, downstream impacts on successful commercialization, and an inclusive innovation corridor across the Mid-South. This effort will advance technology transfer from academic institutions into entrepreneurial ventures that seed emergent, regional ecosystems. The data-driven, performance improvement approach will ensure best practices are evidence-based and create a model for other regions seeking to induce inclusive innovation cluster development.<br/><br/>This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.