Loading…
This event has ended. View the official site or create your own event → Check it out
This event has ended. Create your own
View analytic

Sign up or log in to bookmark your favorites and sync them to your phone or calendar.

Wednesday, March 15
 

8:30am

Tutorial Part I: Essential HPC Finance Practice: Total Cost of Ownership (TCO), Internal Funding, and Cost-Recovery Models
The tutorial provides an impartial, practical, non-sales focused guide to the financial aspects of HPC facilities and service. It presents a rare and invaluable opportunity for HPC managers, practitioners, and stakeholders to learn more about calculating and using TCO models; along with the pros and cons of different internal cost recovery and funding models. Well-managed TCO, return on investment and cost recovery models can be hugely beneficial to HPC managers and operators by demonstrating the value of HPC to the organization, driving the continuation and growth of HPC investment. They can also help uncover practical improvements to deliver better services to users. Attendees will benefit from exploration of the main issues, pros and cons of differing approaches, practical tips, hard-earned experience and potential pitfalls. After the tutorial, attendees will be in a stronger position to calculate and use TCO within their organizations, and to design and use internal cost-recovery models. The tutorial is based on experience across a diverse set of real world cases in various countries, in both private and public sectors, with projects of all sizes and shapes.

Speakers
avatar for Andrew Jones

Andrew Jones

VP HPC Consulting & Services, Numerical Algorithms Group
Originally a scientist using HPC for research, later Research Computing Manager at a UK supercomputer center, now leads consulting/services business. Has led, advised, managed or reviewed the planning and delivery of HPC services and facilities in government, industry and academia, in several countries, including numerous HPC acquistions. Member of the UK government’s e-Infrastructure Leadership Council, advising the government on national... Read More →
DL

Dairsie Latimer

Red Oak Consulting
Dairsie is a highly experienced consultant with a proven track record of successful delivery of complex, multi-million pound IT systems and solutions. Dairsie has worked in a wide variety of roles on supplier side and client side across the commercial and public sectors. Following a career in micro-architecture and software development, he has over ten years' specialist experience in the HPC sector advising on strategy, technology, supporting... Read More →


Wednesday March 15, 2017 8:30am - 10:00am
Room 282

8:30am

Tutorial: An introduction to Intel’s Knights Landing and its Performance Characteristics
Intel’s next generation Xeon Phi, Knights Landing (KNL), brings many changes from the first generation, Knights Corner (KNC). The new processor supports self-­‐‑hosted nodes, connects cores via a mesh topology rather than a ring, and uses a new memory technology, MCDRAM. It is based on Intel’s x86 technology with wide vector units and hardware threads. Many of the lessons learned from using KNC do still apply, such as efficient multi-­‐‑ threading, optimized vectorization, and strided memory access.

In this tutorial we will review the KNL architecture, and discuss the differences between KNC and KNL. We will also discuss the impact of the different MCDRAM memory configurations and the different modes of cluster configuration. Recommendations regarding hybrid MPI+OMP execution when using KNL with the Intel OmniPath fabric will be provided.

We will also analyze the performance of some of the most popular applications in Stampede when running on KNL, and compare it to alternative platforms.


Speakers
CR

Carlos Rosales-Fernandez

Texas Advanced Computing Center, The University of Texas at Austin


Wednesday March 15, 2017 8:30am - 10:00am
Room 280

8:30am

Workshop Part I: Best Practices in Supercomputing Systems Management
This workshop will share Best Practices in Supercomputing Systems Management.
In the last few years, our industry has made great progress improving the reliability of very large MPI jobs in our clusters. Many of the ideas we have implemented came from friends at the Texas Advanced Computing Center and Oak Ridge National Labs. We are organizing a workshop to share best practices in OS deployment and automation, identification of hardware and facilities issues before they impact systems performance and reliability, and systems
instrumentation to improve application performance. Experts from TACC, Oak Ridge, Rice, Chevron and BP will present and share their experience.

Moderators
avatar for Keith Gray

Keith Gray

Manager, High Performance Computing, BP
Keith Gray is Manager of High Performance Computing for BP. The HPC Team supports the computing requirements for BP’s Advanced Seismic Imaging Research efforts. This team supports one of the largest Linux Clusters dedicated to research in Oil and Gas. Mr. Gray graduated from Virginia Tech with a degree in geophysics, and has worked for BP and Amoco since 1985. He was listed in HPCWire’s People to Watch 2006.

Speakers


Wednesday March 15, 2017 8:30am - 10:00am
Room 284

10:00am

Break
Wednesday March 15, 2017 10:00am - 10:30am
Exhibit Hall BRC

10:30am

Tutorial Part II: Essential HPC Finance Practice: Total Cost of Ownership (TCO), Internal Funding, and Cost-Recovery Models
The tutorial provides an impartial, practical, non-sales focused guide to the financial aspects of HPC facilities and service. It presents a rare and invaluable opportunity for HPC managers, practitioners, and stakeholders to learn more about calculating and using TCO models; along with the pros and cons of different internal cost recovery and funding models. Well-managed TCO, return on investment and cost recovery models can be hugely beneficial to HPC managers and operators by demonstrating the value of HPC to the organization, driving the continuation and growth of HPC investment. They can also help uncover practical improvements to deliver better services to users. Attendees will benefit from exploration of the main issues, pros and cons of differing approaches, practical tips, hard-earned experience and potential pitfalls. After the tutorial, attendees will be in a stronger position to calculate and use TCO within their organizations, and to design and use internal cost-recovery models. The tutorial is based on experience across a diverse set of real world cases in various countries, in both private and public sectors, with projects of all sizes and shapes.

Speakers
avatar for Andrew Jones

Andrew Jones

VP HPC Consulting & Services, Numerical Algorithms Group
Originally a scientist using HPC for research, later Research Computing Manager at a UK supercomputer center, now leads consulting/services business. Has led, advised, managed or reviewed the planning and delivery of HPC services and facilities in government, industry and academia, in several countries, including numerous HPC acquistions. Member of the UK government’s e-Infrastructure Leadership Council, advising the government on national... Read More →
DL

Dairsie Latimer

Red Oak Consulting
Dairsie is a highly experienced consultant with a proven track record of successful delivery of complex, multi-million pound IT systems and solutions. Dairsie has worked in a wide variety of roles on supplier side and client side across the commercial and public sectors. Following a career in micro-architecture and software development, he has over ten years' specialist experience in the HPC sector advising on strategy, technology, supporting... Read More →


Wednesday March 15, 2017 10:30am - 12:00pm
Room 282

10:30am

Tutorial: HDF5: An Introduction plus an I/O tuning Case Study
HDF5 is a mature library and file format for the exchange of scientific data, as well as for it's high-performance archival and retrieval. It has emerged as the standard for the underlying architecture in utilities such as pytables and NetCDF. HDF5 bindings exist for python, C++, Fortran and java, with others under active development. Recently ODBC support has been added and interoperability with Apache Spark and other hyperscale technologies has been demonstrated.

HDF5 has deep and rich functionality, and getting started with it can be challenging. This tutorial session will provide an introduction to basic concepts of HDF5, and will showcase how a small subset of its features have been used to tune the performance of a petascale seismic code.

Speakers

Wednesday March 15, 2017 10:30am - 12:00pm
Room 280

10:30am

Workshop Part II: Best Practices in Supercomputing Systems Management
This workshop will share Best Practices in Supercomputing Systems Management.
In the last few years, our industry has made great progress improving the reliability of very large MPI jobs in our clusters. Many of the ideas we have implemented came from friends at the Texas Advanced Computing Center and Oak Ridge National Labs. We are organizing a workshop to share best practices in OS deployment and automation, identification of hardware and facilities issues before they impact systems performance and reliability, and systems
instrumentation to improve application performance. Experts from TACC, Oak Ridge, Rice, Chevron and BP will present and share their experience.

Moderators
avatar for Keith Gray

Keith Gray

Manager, High Performance Computing, BP
Keith Gray is Manager of High Performance Computing for BP. The HPC Team supports the computing requirements for BP’s Advanced Seismic Imaging Research efforts. This team supports one of the largest Linux Clusters dedicated to research in Oil and Gas. Mr. Gray graduated from Virginia Tech with a degree in geophysics, and has worked for BP and Amoco since 1985. He was listed in HPCWire’s People to Watch 2006.

Wednesday March 15, 2017 10:30am - 12:00pm
Room 284

12:00pm

Conference Registration & Networking
Please note, we will have refreshments available from 12-1pm but we are not serving lunch.

Wednesday March 15, 2017 12:00pm - 1:00pm
Room 103 BRC

1:00pm

Welcome
Opening remarks

Speakers
avatar for Jan E. Odegard

Jan E. Odegard

Executive Director, Ken Kennedy Institute for Information Technology, Rice University
Jan E. Odegard Executive Director, Ken Kennedy Institute for Information Technology and Associate Vice President, Research Computing & Cyberinfrastructure at Rice University. Dr. Odegard joined Rice University in 2002, and has over 15 years of experience supporting and enabling research in computing, big-data and information technology. | As Executive Director of the Ken Kennedy Institute, Dr. Odegard co-leads the institute’s mission to... Read More →


Wednesday March 15, 2017 1:00pm - 1:15pm
Room 103 BRC

1:15pm

Keynote: "High Performance Computing and Full Waveform Inversion: ExxonMobil Perspective", John Eastwood, ExxonMobil
PRESENTATION NOT AVAILABLE

3D seismic images underpin almost all geoscience interpretation/analysis for upstream opportunity generation in the upstream oil and gas industry. Full Waveform Inversion (FWI), is a highly compute intensive emerging technology to create images and models of the subsurface.
FWI utilizes the seismic wave propagation equation (up to full physics) in to simulate seismic field records (forward model), based on an initial representation 3D model of geophysical parameters of the subsurface.  FWI iteratively updates the geophysical parameter 3D model to improve the match between the actual field data and simulated data to a user specified level (inversion).
The compute costs and time required to apply FWI can be a limitation to use.    Optimization, of many components of the full system including hardware, software, project selection, level of physics can lower these barriers.
This talk discusses the challenges FWI creates for modern high performance computing systems and possible approaches to full system optimization in order to lower compute costs and improve cycle time.

Speakers
avatar for John Eastwood

John Eastwood

Geophysics Manager, ExxonMobil
John Eastwood is the Geophysics Manager at ExxonMobil for Seismic Imaging/Processing/FWI Research and Applications and Acquisition Research. Previous to this role John has worked for Exxonmobil as a manager in Exploration, Production and Research in both Canada and the United States.  He has a Phd in Geophysics from the University of Alberta. He has worked with the SEG as Secretary Treasurer, The Leading Edge Editor and Finance Committee... Read More →


Wednesday March 15, 2017 1:15pm - 2:00pm
Room 103 BRC

2:00pm

Plenary: "Things to consider – The Changing Landscape of HPC and Datacenter", Debra Goldfarb, Intel
PRESENTATION NOT AVAILABLE

Speakers
avatar for Debra Goldfarb

Debra Goldfarb

Chief Analyst, Senior Director of MI, Senior Principal Engineer, Datacenter Group, Intel
Debra is the Chief Analyst and Senior Director of Market Intelligence for Intel’s Data Center Group.  Debra’s nearly 30 year career in High Performance Computing  started at IDC where she spent 17 years leading the company’s  presence in high end computing in government, academia and industry.  She was instrumental in driving science and technology policy initiatives in the US and abroad, highlighting the... Read More →


Wednesday March 15, 2017 2:00pm - 2:30pm
Room 103 BRC

2:30pm

Plenary: "Scaling Deep Learning Applications: Theoretical and Practical Limits", Janis Keuper, Fraunhofer ITWM

WATCH THE PRESENTATION

Our recent research [1] showed, that distributedly scaling the training of Deep
Neural Networks is a very hard problem which still lacks feasible solutions. In
this talk, we give an introduction to the main theoretical limits prohibiting efficient scalability and current approaches towards applicable solutions.  Also, we discuss many additional practical problems of real world deployments of Deep Learning algorithms to HPC platforms.
Finally, we discuss the consequences and implications of these limitations in the  context of Oil and Gas applications with a focus on algorithms for seismic imaging.

[1] Janis Keuper and Franz-Josef Pfreundt. 2016. Distributed training of deep neural networks: theoretical and practical limits of parallel scalability. In Proceedings of the Workshop on Machine Learning in High Performance Computing Environments (MLHPC '16) at Supercomputing 16.

Speakers
avatar for Janis Keuper

Janis Keuper

Senior Researcher, Frau
Janis Keuper is a Senior Scientist at ITWM with over 10 years research | and application experience in the fields of Machine Learning, Pattern | Recognition and Computer Vision. Before joining ITWM in 2012, he was a | Group Leader at the Intel Visual Computing Institute (Saarbrücken, | Germany). Janis received his Masters and PhD degrees in Computer | Science form the Albert-Ludwigs University in Freiburg and did his... Read More →
avatar for Franz-Josef Pfreundt

Franz-Josef Pfreundt

Division Director, Competence Center for HPC
I studied Mathematics, Physics and Computer Science and got a PhD in Mathematical Physics and helped to found the Fraunhofer ITWM in kaiserslautern/ Germany. Today I am leading the Competence Center for HPC. | In my group we developed the GPI programming model and the BeeGFS parallel file system. With GPI-Space we created a parallel execution and development framework that is used today to steer large scale big data applications and... Read More →



Wednesday March 15, 2017 2:30pm - 3:00pm
Room 103 BRC

3:00pm

Disruptive Technology: Drilling Deep with Machine Learning as an Enterprise Enabled Micro Service
TWO MINUTE LIGHTENING TALK

Numerous data points quantify the current, significant level of interest in Machine Learning - and, when Big Data is involved, Deep Learning. And although Machine Learning is already receiving some degree of uptake in the oil and gas industry, as its value is validated in proof-of-concept initiatives, adoption across the enterprise presents additional opportunities and challenges. Because Machine Learning has the potential to contribute to numerous, existing workloads and workflows to varying degrees, a departure from classic architectural patterns may be warranted. In particular, it is argued that departures that seek to introduce Machine Learning capabilities as a micro service warrant attention. To fix ideas, a specific use case is shared: Machine Learning capabilities from Apache Spark are containerized and scheduled for use via Navops Command by Univa, in a container cluster based on Kubernetes. Starting with use-case examples such as this one, organizations can be better informed and inclined to consider micro services based architectures as they adopt and adapt Machine Learning capabilities into their existing workloads and workflows.

Speakers
avatar for Ian Lumb

Ian Lumb

Solutions Architect, Univa Corporation
As an HPC specialist, Ian Lumb has spent about two decades at the global intersection of IT and science. Ian received his B.Sc. from Montreal's McGill University, and then an M.Sc. from York University in Toronto. Although his undergraduate and graduate studies emphasized geophysics, Ian’s current interests include workload orchestration and container optimization for HPC to Big Data Analytics in clusters and clouds. Ian enjoys discussing... Read More →



Wednesday March 15, 2017 3:00pm - 3:04pm
Room 103 BRC

3:04pm

Disruptive Technology: Global Collaborative E&P Cloud workflow on Amazon
TWO MINUTE LIGHTENING TALK

The combination of advanced cloud infrastructure with modern software architectures can enable complex E&P workflows. Modern data management enables the integration between on premise and cloud interpretation, while browser based remote visualization enables remote users to collaborate on any hardware (PCs, Tablets, etc...).
The scalability and performance optimization that can be achieved on modern clouds with the power of GPUs is multiplied several orders of magnitude by the advanced software architecture of HueSpace to the point where it is worth asking the question:

What if anything of what you are trying to do every day could be 1,000 times faster? Not just an algorithm, but the whole workflow ? How would this change the way you work, what questions could you ask and how it would change the business fundamentals of your domain?

We will demonstrate a live workflow where all the North Sea data is fully available and accessible using Headwave interpretation software. Multiple interpreters can freely collaborate across the globe, sharing interpretations that are uploaded in a common environment and where advanced HPC algorithms are executed in real time to enable interactive decisions making.
This results in modern exploration workflows that contribute to cost effective exploration and related business decisions in the world of $40-50/bbl.
-------
This presentation will be co-presented by Amazon and by Hue, and the live demo environment will be offered for free during the conference and for a week after the conference to all Rice HPC participants (we are validating this, but we will try to do this or at least enable the first 50-100 users...). We would also be interested in enabling some Rice Univ. students to build HPC algorithms on our platform and showcasing it during the workflow, to demonstrate how this environment can open up opportunities for young scientists with entrepreneurial aspirations to build a business.

Speakers
avatar for Michele Isernia

Michele Isernia

VP Strategy & Alliances, Hue Technology N.A
Ideas and Innovation grounded to global business development, mostly in "enterprise" type businesses.


Wednesday March 15, 2017 3:04pm - 3:08pm
Room 103 BRC

3:08pm

Disruptive Technology: IBM
TWO MINUTE LIGHTENING TALKS

Speakers


Wednesday March 15, 2017 3:08pm - 3:12pm
Room 103 BRC

3:12pm

3:16pm

Disruptive Technology: OpenSFS Reloaded
TWO MINUTE LIGHTENING TALK

OpenSFS is a vendor neutral, member supported non-profit organization bringing together the open source file system community for the high performance computing sector. Our mission is to aggregate community resources and be the center of collaborative activities to ensure efficient coordination of technology advancement, development, and education. OpenSFS has recently reorganized into a user-driven organization with flat and low annual membership fees and is looking to engage the open source file system community and increase membership.

Speakers

Wednesday March 15, 2017 3:16pm - 3:20pm
Room 103 BRC

3:20pm

Disruptive Technology: Sustaining an Ecosystem of Open Innovation
TWO MINUTE LIGHTENING TALK

There’s a vital need for IT to be relevant to the business. Technology innovation drives competitive differentiation for enterprises of the future. It takes the trifecta of Business, Technology and Ecosystem to come together to make innovation relevant. This session details the salient characteristics of all three aspects - Business, Technology and Ecosystem leading up to how innovation can be made real in the Red Hat Open Innovation Labs with business context.

Speakers
avatar for E.G. Nadhan

E.G. Nadhan

Chief Technology Strategist, Red Hat
With over 25 years of experience in the IT industry selling, delivering and managing enterprise solutions for global enterprises, E.G.Nadhan is the Chief Technology Strategist at Red Hat (Central Region) working with the executive leadership of enterprises to innovatively drive Digital Transformation with a healthy blend of emerging solutions and a DevOps mindset. Listed in the top 100 Digital Transformation influencers on Twitter, Nadhan also... Read More →


Wednesday March 15, 2017 3:20pm - 3:24pm
Room 103 BRC

3:24pm

Break
Wednesday March 15, 2017 3:24pm - 3:54pm
Exhibit Hall BRC

4:00pm

Algorithms and Performance: Weight-Adjusted Discontinuous Galerkin methods for Acoustic and Elastic Wave Propagation
Discontinuous Galerkin methods allow for high order accurate simulations of wave propagation in heterogeneous media and complex geometries. However, models of media and geometry are typically approximated with low order accuracy, as high order approximations of these models can introduce high storage costs or the loss of stability and high order accuracy.

In this talk, we describe how Weight-adjusted discontinuous Galerkin (WADG) methods allow for high order accurate approximations of heterogeneous media and geometry while preserving low storage costs, energy stability, and high order accuracy. Numerical results confirm the accuracy and computational performance of WADG methods for benchmark problems in acoustic and elastic wave propagation.

Moderators
HC

Henri Calandra

Total
Henri Calandra obtained his M.Sc. in mathematics in 1984 and a Ph.D. in mathematics in 1987 from the Universite des Pays de l’Adour in Pau, France. He joined Cray Research France in 1987 and worked on seismic applications. In 1989 he joined the applied mathematics department of the French Atomic Agency. In 1990 he started working for Total SA. After 12 years of work in high performance computing and as project leader for Pre-stack Depth... Read More →
avatar for Ernesto Prudencio

Ernesto Prudencio

Senior Software Engineer, Schlumberger
Ernesto E. Prudencio combines a BSc in electronics engineering (1990, Brazil), a MSc in applied mathematics (domain decomposition methods, 2001, Brazil), and a PhD in computer science / numerical analysis (PDE-constrained optimization, 2005, Boulder, CO), with professional experience in industry (IBM, Integris, Schlumberger), in national laboratories (ANL, SLAC) and academia (UT Austin). He has been working in Schlumberger since September of... Read More →

Speakers


Wednesday March 15, 2017 4:00pm - 4:20pm
Room 280

4:00pm

Facilities, Infrastructure & Networking: ExSeisPIOL: Extreme-Scale Parallel I/O for Seismic Workflows
Seismic data-sets generated during Oil & Gas exploration are extremely large and are broken
into data files which can be 100s of GiBs to 10s of TiBs and larger. From a software
development perspective, parallel I/O with seismic files is complex and challenging and thus
involves a significant percentage of developer productivity. The fundamental factor for
complexity is the nature of the I/O-access patterns which tend to be varied and involve large
amounts of data, resulting in the sub-optimal exploitation of current and emerging large-scale
HPC platforms.

The primary file formats in use within petroleum seismology have also accumulated 40 years
of legacy. For example, the de-facto standard, SEG-Y, requires fixed byte positions for the
storage of pre-defined trace parameters, uses the lower-precision IBM floating point format
for trace data and also uses EBCDIC for text. SEG-Y also interleaves both data and metadata
in the same file due to its origins in tape storage. The format does not support many common
use cases that geophysicists require since the nature of seismic workflows makes this
difficult in a static file format. Processing the peculiarities of the SEG-Y format can further
increase the complexity of the code base as a result.

In this presentation, we provide an overview of a new high-performance parallel I/O library,
ExSeisPIOL, dedicated to seismic processing workflows for petroleum seismology. Developed
to target SEG-Y, ExSeisPIOL is optimised for the structure of seismic files to deliver significant
improvements in both productivity and performance on large-scale HPC systems. Using
modern OO design, the library is developed on top of an interface abstraction which targets
MPI-I/O and other I/O interfaces. The library also targets the latest in burst buffer technology
with appropriate access semantics or through the direct exploitation of the low-level
interfaces, to deliver a significant improvement to I/O performance over standalone parallel
filesystems such as Lustre and GPFS.

We will describe the ExSeisPIOL in more detail by providing an overview of the multi-layered
structure as well as the two primary APIs which provide a high-level and low-level abstraction.
We will discuss real-world Oil & Gas use cases on large-scale HPC systems using burst-buffer
technology as well as a port to an existing seismic migration code (>15% code reduction).

Moderators
avatar for Keith Gray

Keith Gray

Manager, High Performance Computing, BP
Keith Gray is Manager of High Performance Computing for BP. The HPC Team supports the computing requirements for BP’s Advanced Seismic Imaging Research efforts. This team supports one of the largest Linux Clusters dedicated to research in Oil and Gas. Mr. Gray graduated from Virginia Tech with a degree in geophysics, and has worked for BP and Amoco since 1985. He was listed in HPCWire’s People to Watch 2006.

Speakers


Wednesday March 15, 2017 4:00pm - 4:20pm
Room 103 BRC

4:20pm

Algorithms and Performance: High Performance Low Rank Schur Complement for the Helmholtz Equation
Solving the Helmholtz equation represents an important challenge in large-scale 3D seismic applications. A sparse direct solvers is a method of choice in presence of many right-hand sides on a large domain for a selected set of frequencies. In particular, multifrontal (i.e., MUMPS/SuperLU) and supernodal (i.e., PaStiX) solvers exhibit formally dense Schur complements on the last root separators and the first level blocks, respectively, obtained from nested dissection graph partitioning. These dense Schur complements turn out to operate on data-sparse, low rank structured matrix blocks. Data compression (through SVD, RSVD, RRQR, ACA, etc.) can then occur by means of approximating each underlying tile with a given accuracy threshold. After truncation, the resulting low rank tile data structure corresponds to an outer product of two tall-and-skinny matrices of width k, with k being the rank of the compressed block. The Schur complement computation needs now to take into account the new data structure. There have been many recent works on exploiting the low rankness in direct sparse solvers and preconditioning [Engquist and Ying 2011, Kriemann 2013, Xia 2013, Ambikasaran 2013, Aminfar et. al 2014, Amestoy et. al 2015, etc.].
We design and implement a new efficient Schur complement on massively parallel hardware architectures, such as Intel KNLs and NVIDIA Pascal GPUs. The main idea is to refactor the the Schur complement code by moving from a tile-centric to a kernel-centric variant of the code in order to expose batch kernel executions. The low arithmetic intensity of the numerical kernels (due to very small rank sizes) and the resulting latency overhead can be compensated by increasing the occupancy on the system. This requires the extension of the standard BLAS kernels into batch execution mode as well as variable block sizes to handle the rank size heterogeneity. This new BLAS kernel collection is being aggressively investigated by the community and industrial vendors from an API and performance point of view.
We present the HiCMA library, which performs hierarchical computations on manycore architectures using low rank tile-structured matrix. HiCMA relies on the KBLAS library for performance, which implements highly tuned batch BLAS kernels on NVIDIA GPUs on very small, variable rank sizes. On Intel architecture, HiCMA relies on OpenMP to batch sequential MKL kernels across the processing units. We describe the Schur complement computation for the Helmholtz equation within these aforementioned frameworks. We report the resulting memory footprint, arithmetic complexity, and performance on various hardware architectures and compare against state-of-the-art dense numerical libraries.

Moderators
HC

Henri Calandra

Total
Henri Calandra obtained his M.Sc. in mathematics in 1984 and a Ph.D. in mathematics in 1987 from the Universite des Pays de l’Adour in Pau, France. He joined Cray Research France in 1987 and worked on seismic applications. In 1989 he joined the applied mathematics department of the French Atomic Agency. In 1990 he started working for Total SA. After 12 years of work in high performance computing and as project leader for Pre-stack Depth... Read More →
avatar for Ernesto Prudencio

Ernesto Prudencio

Senior Software Engineer, Schlumberger
Ernesto E. Prudencio combines a BSc in electronics engineering (1990, Brazil), a MSc in applied mathematics (domain decomposition methods, 2001, Brazil), and a PhD in computer science / numerical analysis (PDE-constrained optimization, 2005, Boulder, CO), with professional experience in industry (IBM, Integris, Schlumberger), in national laboratories (ANL, SLAC) and academia (UT Austin). He has been working in Schlumberger since September of... Read More →

Speakers
HL

Hatem Ltaief

Senior Research Scientist, KAUST
High performance computing | Numerical linear algebra | Performance optimization



Wednesday March 15, 2017 4:20pm - 4:40pm
Room 280

4:20pm

Facilities, Infrastructure & Networking: Achieving the Ultimate Efficiency for Seismic Analysis
The pressure to reduce both operating and capital costs in seismic data analysis drives an on-going demand for efficiency improvements in computational processing facilities. This is against a background of dramatically increasing seismic data volumes. Wide/Multi/Rich-azimuth methods using multi-sensor arrays and sophisticated acquisition techniques are producing higher-fidelity subsurface images, and modern analytics techniques are enabling continued advancement in the interpretation of seismic data for both newly acquired data and historical oil field data. As a result of the volume and scale of the seismic data required for modern HPC-based seismic processing and imaging, the performance of the associated storage subsystem can be a source of the greatest overall efficiency improvements.

Seismic analysis is particularly challenging for today’s file systems due to a tendency towards large random IO and share-file IO. Therefore, improving IO efficiencies for complex seismic workloads is key. The benefit of a true parallel file system is the very high single-client performance that can be delivered and sustained even when many hundreds of clients are working concurrently

This session will share the results experimental benchmarks to attain optimal IO rates for Paradigm’s Echos application workloads.

Moderators
avatar for Keith Gray

Keith Gray

Manager, High Performance Computing, BP
Keith Gray is Manager of High Performance Computing for BP. The HPC Team supports the computing requirements for BP’s Advanced Seismic Imaging Research efforts. This team supports one of the largest Linux Clusters dedicated to research in Oil and Gas. Mr. Gray graduated from Virginia Tech with a degree in geophysics, and has worked for BP and Amoco since 1985. He was listed in HPCWire’s People to Watch 2006.

Speakers


Wednesday March 15, 2017 4:20pm - 4:40pm
Room 103 BRC

4:40pm

Algorithms and Performance: Improving Scalability and Performance of Linear System Solves in Pore-scale Simulations
Pore-scale fluid simulation is an integral part of hydrocarbon exploration and production. The numerical solution of the Cahn-Hilliard equation, which governs the separation of a two-component fluid mixture, is an essential and computationally demanding part of our simulations. In this talk, we present two approaches to improve the performance and scalability of the solution of the large, sparse linear systems arising from the equation discretization, since these system solves are the main performance bottleneck of the simulation.

The iterative solvers used to solve these systems rely on fast sparse matrix-vector products (SpMVs), and the size of the problems require the distribution of the linear systems over multiple computing nodes. Thus, a fully asynchronous, hybrid parallel SpMV implementation is introduced, which overlaps computation and communication in order to achieve high scalability.

Our second approach to improve the performance of the linear systems solves is to exploit hierarchical scale separation (HSS), a recently developed multigrid-type scheme to solve linear systems arising from discontinuous Galerkin methods. HSS profits from the spatial locality of the numerical solution by splitting each linear system into a coarse-scale system of reduced size and a set of very small, decoupled fine-scale systems. For the former, a standard iterative solver is employed, while the latter are solved with a direct solver. This splitting improves the overall performance of the linear system solves, since the iterative solver benefits from the reduced system size, while the direct solves of the decoupled systems can be parallelized very efficiently. We also show a modification of the HSS algorithm that further improves scalability and performance.

Both approaches help improving the performance of linear systems solves in our application. The asynchronous SpMV implementation displays high scalability, and the modified HSS algorithm greatly accelerates the GMRES solve with its multigrid-type approach. Experimental results from the application of the above techniques to the Cahn-Hilliard equation are presented.

Moderators
HC

Henri Calandra

Total
Henri Calandra obtained his M.Sc. in mathematics in 1984 and a Ph.D. in mathematics in 1987 from the Universite des Pays de l’Adour in Pau, France. He joined Cray Research France in 1987 and worked on seismic applications. In 1989 he joined the applied mathematics department of the French Atomic Agency. In 1990 he started working for Total SA. After 12 years of work in high performance computing and as project leader for Pre-stack Depth... Read More →
avatar for Ernesto Prudencio

Ernesto Prudencio

Senior Software Engineer, Schlumberger
Ernesto E. Prudencio combines a BSc in electronics engineering (1990, Brazil), a MSc in applied mathematics (domain decomposition methods, 2001, Brazil), and a PhD in computer science / numerical analysis (PDE-constrained optimization, 2005, Boulder, CO), with professional experience in industry (IBM, Integris, Schlumberger), in national laboratories (ANL, SLAC) and academia (UT Austin). He has been working in Schlumberger since September of... Read More →

Speakers


Wednesday March 15, 2017 4:40pm - 5:00pm
Room 280

4:40pm

Facilities, Infrastructure & Networking: A Case Study of the Impact of System Profiling in a Seismic Processing Data Center
An effort began three years ago to profile various types of equipment in CGG’s Houston data center to better understand the composite workload. This paper gives an overview of the effort as well as specific examples of how it has led to improved equipment selection. The paper begins with a sketch of the Houston data center and the main types of equipment. Two types of storage system profiles are covered followed by two types of GPU system profiles. The study concludes with two types of CPU workload profiles. In each of the subsystem categories the profiling effort has resulted in improved equipment selection (HW), improved utilization, or deployment of new technologies. In most of these cases profiling has resulted in an increase of Performance/$ by 25% or more. A common theme of these improvements is to shift the data center design point from “+3 standard deviations” to “+1 standard deviation” with respect to workload requirements.

Moderators
avatar for Keith Gray

Keith Gray

Manager, High Performance Computing, BP
Keith Gray is Manager of High Performance Computing for BP. The HPC Team supports the computing requirements for BP’s Advanced Seismic Imaging Research efforts. This team supports one of the largest Linux Clusters dedicated to research in Oil and Gas. Mr. Gray graduated from Virginia Tech with a degree in geophysics, and has worked for BP and Amoco since 1985. He was listed in HPCWire’s People to Watch 2006.

Speakers


Wednesday March 15, 2017 4:40pm - 5:00pm
Room 103 BRC

5:00pm

Algorithms and Performance: An Efficient and High Order Accurate Solution Technique for Scattering in Variable Media
PRESENTATION NOT AVAILABLE

Classic numerical partial differential equation techniques face two big
problems when applied to problems with highly oscillatory solutions. First there is the so-called ``pollution'' (dispersion) error, demanding an increasing number of degrees of freedom per wavelength in order to maintain fixed accuracy as wavenumber grows. Second, iterative solvers that are typically used to solve the linear system resulting from discretization are slow to converge due to
ill-conditioning.

While there is much on going work to resolving these problems, this talk presents an alternative solution technique for high frequency scattering problems in variable media. This method does not observe pollution and has an efficient direct solver is called the Hierarchical Poincaré-Steklov (HPS) scheme. The technique uses a classical spectral collocation method on a collection of disjoint leaf
boxes whose union is the domain. On each leaf box approximate solution
operators and Poincaré-Steklov operators such as Dirichlet-to-Neumann operators are
constructed. Then boxes are ``glued'' together in a hierarchical fashion creating
approximate solution and Poincaré-Steklov operators for the union of two boxes.
Once this precomputation is complete, new boundary conditions and source functions
can be processed by applying the solution operators via a collection of small matrix vector multiplies. The resulting method has computational cost that is asymptotically the same as the nested dissection method but has high order accuracy even for problems with highly oscillatory solutions. For example when applied to the Helmholtz problem with a fixed twelve points per wavelength, the HPS method achieves eight digits of accuracy.

In addition to presenting a high level view of the HPS method, we will illustrate
it performance in both the forward and inverse scattering setting. Next, the
adaptive version of the method will be present with numerical results to report
on its performance.

Moderators
HC

Henri Calandra

Total
Henri Calandra obtained his M.Sc. in mathematics in 1984 and a Ph.D. in mathematics in 1987 from the Universite des Pays de l’Adour in Pau, France. He joined Cray Research France in 1987 and worked on seismic applications. In 1989 he joined the applied mathematics department of the French Atomic Agency. In 1990 he started working for Total SA. After 12 years of work in high performance computing and as project leader for Pre-stack Depth... Read More →
avatar for Ernesto Prudencio

Ernesto Prudencio

Senior Software Engineer, Schlumberger
Ernesto E. Prudencio combines a BSc in electronics engineering (1990, Brazil), a MSc in applied mathematics (domain decomposition methods, 2001, Brazil), and a PhD in computer science / numerical analysis (PDE-constrained optimization, 2005, Boulder, CO), with professional experience in industry (IBM, Integris, Schlumberger), in national laboratories (ANL, SLAC) and academia (UT Austin). He has been working in Schlumberger since September of... Read More →

Speakers

Wednesday March 15, 2017 5:00pm - 5:20pm
Room 280

5:00pm

Facilities, Infrastructure & Networking: Building an Internal Cloud
Moderators
avatar for Keith Gray

Keith Gray

Manager, High Performance Computing, BP
Keith Gray is Manager of High Performance Computing for BP. The HPC Team supports the computing requirements for BP’s Advanced Seismic Imaging Research efforts. This team supports one of the largest Linux Clusters dedicated to research in Oil and Gas. Mr. Gray graduated from Virginia Tech with a degree in geophysics, and has worked for BP and Amoco since 1985. He was listed in HPCWire’s People to Watch 2006.

Speakers


Wednesday March 15, 2017 5:00pm - 5:20pm
Room 103 BRC

5:20pm

 
Thursday, March 16
 

7:30am

Conference Registration, Breakfast & Networking
Thursday March 16, 2017 7:30am - 8:30am
Exhibit Hall BRC

8:30am

Welcome
Speakers
avatar for Jan E. Odegard

Jan E. Odegard

Executive Director, Ken Kennedy Institute for Information Technology, Rice University
Jan E. Odegard Executive Director, Ken Kennedy Institute for Information Technology and Associate Vice President, Research Computing & Cyberinfrastructure at Rice University. Dr. Odegard joined Rice University in 2002, and has over 15 years of experience supporting and enabling research in computing, big-data and information technology. | As Executive Director of the Ken Kennedy Institute, Dr. Odegard co-leads the institute’s mission to... Read More →


Thursday March 16, 2017 8:30am - 8:45am
Room 103 BRC

8:45am

Keynote: "Algorithmic Adaptations to Extreme Scale" David Keyes, King Abdullah University

Algorithmic adaptations to use next-generation computers closer to their potential are underway in Oil & Gas and many other fields. Instead of squeezing out flops – the traditional goal of algorithmic optimality, which once served as a reasonable proxy for all associated costs – algorithms must now squeeze synchronizations, memory, and data transfers, while extra flops on locally cached data represent only small costs in time and energy. After decades of programming model stability with bulk synchronous processing, new programming models and new algorithmic capabilities (to make forays into, e.g., inverse problems, data assimilation, and uncertainty quantification) must be co-designed with the hardware. We briefly recap the architectural constraints, then concentrate on two kernels that each occupy a large portion of all scientific computing cycles: large dense symmetric/Hermitian systems (covariances, Hamiltonians, Hessians, Schur complements) and large sparse Poisson/Helmholtz systems (solids, fluids, electromagnetism, radiation diffusion, gravitation).  We examine progress in porting solvers for these kernels (e.g., fast multipole, hierarchically low rank matrices, multigrid) to the hybrid distributed-shared programming environment, including the GPU and the MIC architectures.


Speakers
avatar for David Keyes

David Keyes

Director, Extreme Computing Research Center, King Abdullah University of Science and Technology
David Keyes is the director of the Extreme Computing Research Center at King Abdullah University of Science and Technology, where he was a founding dean in 2009, and an adjoint professor of applied mathematics at Columbia University. Keyes earned his BSE in Aerospace and Mechanical Engineering from Princeton and his PhD in Applied Mathematics from Harvard. He works at the algorithmic interface between parallel computing and the... Read More →



Thursday March 16, 2017 8:45am - 9:30am
Room 103 BRC

9:30am

Plenary: "Big Compute: Under the Hood", Alan Lee, AMD
PRESENTATION NOT AVAILABLE.

Speakers
avatar for Alan Lee

Alan Lee

CVP Research & Advanced Development, AMD Research
Alan Lee is the Corporate Vice President of Research and Advanced Development at AMD and the President and CEO of AMD Advanced Research LLC, the division of AMD responsible for external research contracts.  Alan is the founder of AMD Research, and he currently leads hundreds of elite engineers across AMD’s international sites. |   | Prior to joining AMD, Alan was CEO of a privately-held company specializing in financial... Read More →


Thursday March 16, 2017 9:30am - 10:00am
Room 103 BRC

10:00am

Plenary: "Data Center Impacts from the Convergence of High Performance and Cognitive Computing", Jim Kahle, IBM

We describe the coming convergence of traditional high performance computing and emerging cognitive computing workloads and discuss impacts to systems architecture and to the data center. To illustrate the opportunities and challenges, we provide some examples of applications in collaborative development in science domains and describe a proof-of-concept effort which deploys a container based high performance computing and cognitive computing software stack in IBM Research.


Speakers
avatar for Jim Kahle

Jim Kahle

CTO and Chief Architect for Data Centric Deep Computing Systems, IBM
Jim Kahle is a graduate of Rice University, and for more than 30 years at IBM he has held numerous managerial and technical positions. He is a renowned expert in the microprocessor industry, achieving the distinction of IBM Fellow, and currently is the CTO and Chief Architect for Data Centric Deep Computing systems. Previously he was Chief Technical lead for Power 7. Before that, Jim led the Collaborative design for Cell which was a partnership... Read More →



Thursday March 16, 2017 10:00am - 10:30am
Room 103 BRC

10:30am

Break
Thursday March 16, 2017 10:30am - 11:00am
Exhibit Hall BRC

11:00am

Programming: Revisiting Wave Propagation Applications on GPUs: Improving the Accuracy and Performance Tradeoff and Implementing Portability via OCCA
We recently revisited our finite-difference (FD) wave propagation codes to integrate hardware and software portability and to improve computational performance and software reusability. Previously, we developed our RTM code using CUDA in a joint effort with Nvidia to design computationally efficient GPU kernels for the FD schemed used for acoustic wave propagation. This previous design (discussed at past Rice O&G HPC conferences) was geared for the first generation of Nvidia Tesla GPUs which did not have the ability to carry out peer-to-peer or overlapped data transfers from device-to-device or device-to-host, respectively. Furthermore, newer generations of GPUs have larger register and shared memory space, as well as better and less stringent global memory accessing abilities. Another caveat was that these kernels were written in CUDA which meant that we didn’t have the option to run these kernels on other architectures such as CPUs, non-Nvidia GPUs, Xion Phi, and other accelerating coprocessors. Our previous software design was also targeted for RTM, which didn’t easily allow for the portability of our wave propagators to be used for newer seismic imaging and velocity model building techniques such as least squares RTM (LSRTM) (and others) and full waveform inversion (FWI). Finally, we improved our numerics by incorporating variable optimized FD coefficients into the wave propagators.

In this talk, we will discuss our software design strategy including our use of OCCA to implement FD wave propagators with hardware and software portability. We will also discuss specific implementations that led to improved performance and accuracy including the use of better numerics. Furthermore, we will compare performance results between various generations of Nvidia GPUs and between isotropic, VTI, and TTI wave propagators.

Moderators
avatar for Alex Loddoch

Alex Loddoch

TechExpert HPC, Chevron
avatar for Scott Morton

Scott Morton

Manager and Global Geophysical Advisor, Hess Corporation
Scott Morton has 25 years of experience in computational and theoretical physics distributed between academia, the computer industry and the petroleum industry. Although originally trained as an astrophysicist, he switched to geophysics when he joined Shell in 1991 to do research and development in seismic imaging. Scott spent the next 7 years distributed between Shell, Thinking Machines, Cray Research and SGI, gaining expertise in both... Read More →

Speakers
avatar for Thomas Cullison

Thomas Cullison

Solutions Architect Oil & Gas, NVIDIA
Thomas is a Solutions Architect at NVIDIA with a focus on the Oil & Gas industry, and he has a background in computational geophysics, algorithms, and development of HPC and DSP applications for seismic imaging and seismic data processing. As an undergraduate at Colorado School of Mines (CSM), Thomas double majored in geophysical engineering and computer science, and as a graduate, he received a MSc degree in mathematical and computer... Read More →



Thursday March 16, 2017 11:00am - 11:20am
Room 103 BRC

11:20am

Programming: Leveraging Symbolic Math for Rapid Development of Applications for Seismic Modeling
Wave propagation kernels are the core of many commonly used algorithms for inverse problems in exploration geophysics. While they are easy to write and analyze for the simplified cases, the code quickly becomes complex when the physics needs to be made more precise or the performance of these codes is to be optimized. Significant effort is repeated every time new forms of physics need to be implemented, or a new computing platform to be supported. The use of symbolic mathematics as a domain specific language (DSL) for input, combined with automatic generation of high performance code customized for the target hardware platform is a promising approach to maximize code reuse. Devito is a DSL for finite difference that uses symbolic mathematics to generate optimized code for wave propagation based on a provided wave equation. It enables rapid application development in a field where the average time spent on development has historically been in weeks and months. The Devito DSL system is completely wrapped within the Python programming language and the fact that the running code is in C is completely transparent, making it simple to include Devito as part of a larger workflow including multiple applications over a large cluster.

Moderators
avatar for Alex Loddoch

Alex Loddoch

TechExpert HPC, Chevron
avatar for Scott Morton

Scott Morton

Manager and Global Geophysical Advisor, Hess Corporation
Scott Morton has 25 years of experience in computational and theoretical physics distributed between academia, the computer industry and the petroleum industry. Although originally trained as an astrophysicist, he switched to geophysics when he joined Shell in 1991 to do research and development in seismic imaging. Scott spent the next 7 years distributed between Shell, Thinking Machines, Cray Research and SGI, gaining expertise in both... Read More →

Speakers
avatar for Navjot Kukreja

Navjot Kukreja

Imperial College, London



Thursday March 16, 2017 11:20am - 11:40am
Room 103 BRC

11:40am

Programming: A Pipeline Implementation of One-way Wave Equation Migration
PRESENTATION NOT AVAILABLE

This abstract proposes a pipeline implementation of the one-way wave equation depth migration, in which depths are distributed evenly among computing nodes so that the machine memory is big enough to hold the portion of the image volume on each node. The computing nodes are connected in depth series, where the extrapolated wavefields at the last depth step of one node are sent to the next node as the initial input wavefields to start its downward continuation. Once the pipeline is filled up, this implementation allows overlapping the one-way wave equation migration among nodes with synchronization via MPI point-to-point communication between neighbor nodes. The intent is to mitigate the burden of the I/O and network traffic when producing large image gathers with one-way wave equation migration.

Moderators
avatar for Alex Loddoch

Alex Loddoch

TechExpert HPC, Chevron
avatar for Scott Morton

Scott Morton

Manager and Global Geophysical Advisor, Hess Corporation
Scott Morton has 25 years of experience in computational and theoretical physics distributed between academia, the computer industry and the petroleum industry. Although originally trained as an astrophysicist, he switched to geophysics when he joined Shell in 1991 to do research and development in seismic imaging. Scott spent the next 7 years distributed between Shell, Thinking Machines, Cray Research and SGI, gaining expertise in both... Read More →

Speakers

Thursday March 16, 2017 11:40am - 12:00pm
Room 103 BRC

12:00pm

Programming: Many-core Implementation of Numerical 3D Isotropic Acoustic Wave Equation and Analysis of its Performance Portability
I discuss how to tune code in a way that a single source file will perform well across many-core architectures such as Xeon Phi and multi-core architectures such as Xeon. I focus on a common Oil&Gas kernel: the 3D isotropic acoustic wave equation. I give performance results for this kernel across multiple architectures and give strategies on how that performance may be achieved from a single source file. I also discuss an alternative propagation model: first-order forms of the wave equation with PML boundary conditions.

Moderators
avatar for Alex Loddoch

Alex Loddoch

TechExpert HPC, Chevron
avatar for Scott Morton

Scott Morton

Manager and Global Geophysical Advisor, Hess Corporation
Scott Morton has 25 years of experience in computational and theoretical physics distributed between academia, the computer industry and the petroleum industry. Although originally trained as an astrophysicist, he switched to geophysics when he joined Shell in 1991 to do research and development in seismic imaging. Scott spent the next 7 years distributed between Shell, Thinking Machines, Cray Research and SGI, gaining expertise in both... Read More →

Speakers
avatar for Reid Atcheson

Reid Atcheson

Accelerator Software Engineer, Numerical Algorithms Group
Reid consults on High Performance Computing (HPC) with focus on accelerators such as Xeon Phi and Graphics Processing Units. Reid also works with highly distributed code by way of MPI. He has worked in scientific computation throughout his career beginning with Ph.D research on numerical wave propagation and moving into Oil & Gas sector, then finally transitioning into his current role at Numerical Algorithms Group (NAG Inc.) where he works... Read More →



Thursday March 16, 2017 12:00pm - 12:20pm
Room 103 BRC

12:20pm

Break
Lunch & Networking

Thursday March 16, 2017 12:20pm - 1:30pm
Exhibit Hall BRC

1:30pm

Plenary: "The Future of Large Scale IO: Opportunities & Obstacles", Peter Braam, Campaign Storage
 Architectural approaches for storage software face new
challenges as memory and storage devices come closer.  Key issues
include dealing with 100 GB/s local data rates, nano-second latency
and different name- and address-spaces, while also integrating
traditional tiers of storage.  We will explore a few of the
fundamental questions and lingering obstacles, and then visit several
approaches expressed by multiple organizations to deliver fundamental
changes in IO software.

Speakers
avatar for Peter Braam

Peter Braam

CEO, Campaign Storage, LLC.
Peter Braam is a scientist and entrepreneur focused on large | scale computing problems.  After obtaining a PhD in mathematics under | Michael Atiyah, he was an academic at several universities including | Oxford, CMU and Cambridge.  One of his startup companies developed the | Lustre file system which is widely used. Most other products he | designed were sold to major corporations.  From 2013, Peter has been | assisting computing... Read More →



Thursday March 16, 2017 1:30pm - 2:00pm
Room 103 BRC

2:00pm

Plenary: "Supercomputing: Yesterday, Today, and Tomorrow" Peter Ungaro, Cray

Historically, innovations in supercomputing have been eagerly adopted and leveraged by the O&G industry and the industry is poised to take advantage of successful exascale-class systems coming in the future. This talk will recap the last 10 years of supercomputing, highlighting major accomplishments, milestones, and technological advances. The talk will then speculate on what might be coming over the next 10 years as the supercomputing landscape continues to grow and change.  We will look at the future in the context of supercomputing’s growing importance as a key component of competitive business practices, covering new and emerging technologies and also how supercomputing is enabling next generation applications and workflows.



Thursday March 16, 2017 2:00pm - 2:30pm
Room 103 BRC

2:30pm

Panel Introduction
Speakers
avatar for John Mellor-Crummey

John Mellor-Crummey

Professor, Computer Science, Rice University
John Mellor-Crummey’s research focuses on software for high performance and parallel computing, including compilers, tools, and runtime libraries for multicore processors and scalable parallel systems. His current work is principally focused on performance tools for scalable parallel systems. His group’s HPCToolkit performance tools are used on systems around the world ranging from laptops to supercomputers. Recently, he has been... Read More →


Thursday March 16, 2017 2:30pm - 2:45pm
Room 103 BRC

2:45pm

Panel Discussion: High Performance Computing for Oil and Gas: Past, Present, and the Decade Ahead
Moderators
avatar for John Mellor-Crummey

John Mellor-Crummey

Professor, Computer Science, Rice University
John Mellor-Crummey’s research focuses on software for high performance and parallel computing, including compilers, tools, and runtime libraries for multicore processors and scalable parallel systems. His current work is principally focused on performance tools for scalable parallel systems. His group’s HPCToolkit performance tools are used on systems around the world ranging from laptops to supercomputers. Recently, he has been... Read More →

Speakers
avatar for Peter Braam

Peter Braam

CEO, Campaign Storage, LLC.
Peter Braam is a scientist and entrepreneur focused on large | scale computing problems.  After obtaining a PhD in mathematics under | Michael Atiyah, he was an academic at several universities including | Oxford, CMU and Cambridge.  One of his startup companies developed the | Lustre file system which is widely used. Most other products he | designed were sold to major corporations.  From 2013, Peter has been | assisting computing... Read More →
avatar for Debra Goldfarb

Debra Goldfarb

Chief Analyst, Senior Director of MI, Senior Principal Engineer, Datacenter Group, Intel
Debra is the Chief Analyst and Senior Director of Market Intelligence for Intel’s Data Center Group.  Debra’s nearly 30 year career in High Performance Computing  started at IDC where she spent 17 years leading the company’s  presence in high end computing in government, academia and industry.  She was instrumental in driving science and technology policy initiatives in the US and abroad, highlighting the... Read More →
avatar for Jim Kahle

Jim Kahle

CTO and Chief Architect for Data Centric Deep Computing Systems, IBM
Jim Kahle is a graduate of Rice University, and for more than 30 years at IBM he has held numerous managerial and technical positions. He is a renowned expert in the microprocessor industry, achieving the distinction of IBM Fellow, and currently is the CTO and Chief Architect for Data Centric Deep Computing systems. Previously he was Chief Technical lead for Power 7. Before that, Jim led the Collaborative design for Cell which was a partnership... Read More →
avatar for David Keyes

David Keyes

Director, Extreme Computing Research Center, King Abdullah University of Science and Technology
David Keyes is the director of the Extreme Computing Research Center at King Abdullah University of Science and Technology, where he was a founding dean in 2009, and an adjoint professor of applied mathematics at Columbia University. Keyes earned his BSE in Aerospace and Mechanical Engineering from Princeton and his PhD in Applied Mathematics from Harvard. He works at the algorithmic interface between parallel computing and the... Read More →
avatar for Alan Lee

Alan Lee

CVP Research & Advanced Development, AMD Research
Alan Lee is the Corporate Vice President of Research and Advanced Development at AMD and the President and CEO of AMD Advanced Research LLC, the division of AMD responsible for external research contracts.  Alan is the founder of AMD Research, and he currently leads hundreds of elite engineers across AMD’s international sites. |   | Prior to joining AMD, Alan was CEO of a privately-held company specializing in financial... Read More →
avatar for Gilad Shainer

Gilad Shainer

Gilad Shainer has served as Mellanox's vice president of marketing since March 2013. Previously, Mr. Shainer was Mellanox's vice president of marketing development from March 2012 to March 2013. Mr. Shainer joined Mellanox in 2001 as a design engineer and later served in senior marketing management roles between July 2005 and February 2012. Mr. Shainer holds several patents in the field of high-speed networking and contributed to the PCI-SIG... Read More →
avatar for Pete Ungaro

Pete Ungaro

CEO, Cray
Peter Ungaro is president and chief executive officer of Cray Inc., the supercomputer company. He was named CEO of the Year for 2006 by Seattle Business Monthly magazine and one of the “40 under 40” by Corporate Leader Magazine in 2008. In 2013, Bloomberg named him #4 of their “Top Tech Turnaround Artists” for generating shareholder return over 361% since becoming Cray’s CEO. Previously, Ungaro was VP of sales for... Read More →


Thursday March 16, 2017 2:45pm - 4:00pm
Room 103 BRC

4:00pm

Poster: Lithospheric Foundering and Underthrusting beneath Tibet revealed by Adjoint Tomography
New adjoint tomographic images unveil a large-scale, high wave speed structure beneath South-Central Tibet in the middle to lower portions of the upper mantle. We interpret this structure as a remnant of an earlier lithospheric foundering event. Spatial correlations between foundering lithosphere and ultrapotassic and adakitic magmatism support the hypothesis of convective removal of thickened Tibetan lithosphere causing a major rise of Southern Tibet during the Oligocene. Lithospheric foundering induces an asthenospheric drag force, which drives continued underthrusting of the Indian continental lithosphere and associated shortening of the remaining Tibetan lithosphere. We speculate that more recent asthenospheric upwelling leads to a thermal modification of thickened lithosphere beneath Northern Tibet and subsequent surface uplift, consistent with the correlation of recent potassic volcanism and an imaged narrow low wave speed zone in the uppermost mantle. In contrast, the unusually high seismic wave speeds in the uppermost mantle beneath Southern Tibet, reminiscent of images beneath the North American craton, suggest a possible prototype of modern craton formation due to continued under-accretion of Indian continent.


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Statistical Inference of Reticulate Phylogenies and Gene Trees from Multi-locus Data
Evolutionary history explains how species diverged and how genes and traits evolved. Phylogenetic trees have been considered as the basic structure to represent evolutionary relationships. However, the evolutionary history of a set of genomes could be reticulate due to the occurrence of hybridization or horizontal gene transfer. My research focuses on developing novel statistical models and computational methods for inferring reticulate evolutionary histories and studying their mathematical properties. Here I report on the first Bayesian method for sampling the species phylogeny, gene trees, divergence times, and population sizes, from DNA sequences of multiple independent loci. As different numbers of reticulation events correspond to different dimensions in the searching space, I have devised reversible-jump Markov chain Monte Carlo (RJMCMC) techniques for sampling the posterior distribution from genome sequences directly. I have implemented these methods in the publicly available, open-source software package PhyloNet. I demonstrate the utility of the method by analyzing simulated data and reanalyzing three biological data sets. The results demonstrate the significance of not only co-estimating species phylogenies and gene trees, but also accounting for gene flow and ILS simultaneously. My research have provided a big step toward putting networks on equal footing with trees as the model of choice for biologists to use in genomic data analysis.

Speakers

Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: A Discontinuous Galerkin Method for the Direct Numerical Simulation of Flow on Porous Medium
Variations of the Navier-Stokes equation are standard models for the description of viscous liquid and gas flow, used in many industrial fields such as automobile, aerospace, and hydrocarbon production industries. An application in the latter is the direct simulation of fluid transport through porous media at the pore scale, more precisely, on spatial domains that resolve the geometry of porous matrices of rocks. This poster presents a discontinuous Galerkin (DG) method for the Navier-Stokes equation with mass balance coupling defined on voxel sets representing the pore space of rock samples at micrometer scale. Numerical validation tests show optimal convergence rates for the DG discretization indicating the correctness of the numerical scheme. The results of permeability upscaling for one-component single-phase flow and real porous media simulations for two-component flow demonstrate the consistency of the velocity field and mass distribution obtained within our framework and exhibit the potential for tackling realistic problems.

Speakers
avatar for Beatrice Riviere

Beatrice Riviere

Noah Harding Chair and Professor, Rice University


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: A Fast Direct Solver for Elliptic PDEs on Locally-Perturbed Domains
Many problems in science and engineering can be formulated as integral equations with elliptic kernels. In particular, in optimal control and design problems, the domain geometry evolves and results in a sequence of discretized linear systems to be constructed and inverted. While the systems can be constructed and inverted independently, the computational cost is relatively high. In the case where the change in the domain geometry for each new problem is only local, i.e. the geometry remains the same except within a small subdomain, we are able to reduce the cost of inverting the new system by reusing the pre-computed fast direct solvers of the original system. The resulting solver only requires inexpensive matrix-vector multiplications, thus dramatically reducing the cost of inverting the new linear system. Numerical results will illustrate the performance of the solver.


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: A High Order Accurate Direct Solution Technique for High Frequency Problems with Body Loads
Differential equations with highly oscillatory arise in many areas of
science and engineering including geophysics, material science, fluid
dynamics, and medical imaging. When solutions are highly oscillatory,
classic discretization techniques suffer from dispersion and poor
conditioning, which result in accuracy issues and slow convergence for
iterative solvers. This poster presents the recently developed
Hierarchical Poincare-Steklov (HPS) method. The HPS method is a high
order discretization technique that provides accurate solutions even in
the high frequency regime. For example, a problem with 64 wavelengths
per side was solved to 8 digits of accuracy with 16 points per wavelength.
Additionally, the method comes with a direct (as opposed to iterative) solver
that processes solves with nearly linear cost with respect to the number of
discretization points, avoiding the poor performance of iterative solvers
for highly oscillatory solutions. For a test problem with more than 4
million unknowns, after precomputation, the solve takes less than 5
seconds on a modest desktop computer. Since solves are so inexpensive,
the method is ideal for problems with many right hand sides, such as occur
in most applications of interest. Numerical results will illustrate the
performance of the method for a variety of test problems.


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: A Policy-Based System for Dynamic Scaling of Virtual Machine Memory Reservations
Server consolidation via virtualization has become a mainstay of cloud computing technologies. Virtualization enables cloud providers to more efficiently allocate and utilize physical computing resources by decoupling the virtual machines’ resource demands from the physical machine’s resources. Virtual machine performance and cloud provider cost efficiency both depend heavily on the resource allocation policies of the virtualization system. However, current policies for allocating memory are relatively static, constrained by the virtual machines’ initial configurations. As a result, system-wide memory utilization is often sub-optimal, leading to unnecessary paging and performance degradation. This research presents a system for dynamically allocating memory at runtime using six novel dynamic allocation policies. The system allows virtual machines to expand and contract according to their changing demands by uniquely integrating mechanisms such as memory ballooning, memory hotplug, and hypervisor paging. Furthermore, the system provides fairness by guaranteeing each virtual machine a minimum reservation, charging for rentals beyond this minimum, and enforcing timely reclamation of memory.

Speakers

Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Advances in Tetrahedral Discontinuous Galerkin Methods for Wave Propagation
Discontinuous Galerkin (DG) methods on tetrahedral meshes provide geometrically flexible and high order accurate solvers for acoustic and elastic wave propagation. However, compared to Spectral element methods, DG methods are less efficient and less flexible in their representation of velocity models. We present recent results for DG methods on tetrahedral meshes addressing each of these issues.

Speakers

Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Applying Deep Learning on a GPU-based Cluster for Seismic Volume Interpretation with Spark and TensorFlow
In bringing data analytics technology to bear on the problem of computer-assisted interpretation of seismic volumes, the Seismic Analytics Cloud project conducted at Prairie View A&M University has employed the deep learning models in seismic interpretation. In our initial use cases, we train our deep learning models for geological fault identification using labeled fault and non-fault regions in both synthetic and field-recorded data volumes. The computational demands of techniques such as Convolutional Neural Networks (CNN) extend to multiple days on our small-sized clusters for even modest-sized volume analytics problems. This excessive turnaround is mitigated through the use of GPU-based systems using Apache Spark and Google TensorFlow deep learning software. We will show our performance improvement in applying deep learning models on GPU-based clusters. Moreover, we will present how we speed up the Apache Spark performance by bringing two parallel programming models Spark and OpenMP together to perform the deep learning based solutions for seismic interpretation.

We acknowledge the support of the National Science Foundation for this project through a number of research and innovation grant programs.

Speakers
LH

Lei Huang

Assistant Professor, Prairie View A&M University


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Computational Modeling of Electro-mechanical and Catalytic Properties of Materials
The properties of materials can be effectively influenced by applied stress and defects. We explore the many-body effects on the electronic properties of biaxial strained monolayer MoS2 and WS2, and the effects of uniaxial stress along an arbitrary direction on mechanical and electronic properties of phosphorene. Various perturbative corrections to the density functional theory (DFT) electronic structure, e.g. GW, spin-orbit coupling, as well as many-body excitonic and trionic effects are considered, and accurate band gaps as a function of homogeneous biaxial strain in MoS2 and WS2 are calculated. All of these corrections are shown to be of comparable magnitudes and need to be included in order to obtain an accurate electronic structure. The effects of uniaxial stress on mechanical and electronic properties of phosphorene show the enhancement of inherent anisotropy. Basic physical quantities including Young’s modulus, Poisson’s ratio, band gap, and effective carrier masses under external stress are all computed from first principles using DFT, while the final results are presented in compact analytical forms. We also reveal the catalytic mechanism of the defected SnS {131} facet and defected Bi2S3 {211} facet in iodine reduction reaction (IRR). DFT results demonstrate that their catalytic activity are promoted by vacancies defects. These computational modeling can provide suggestions for experimental research.


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Convex Optimization for High-Dimensional Portfolio Construction
Investors in financial markets face a series of complex trade-offs in
constructing an optimal investment portfolio. For example, an investor
might seek to simultaneously maximize his expected return and minimize
his risk, while also diversifying across asset classes, minimizing his
transaction costs, and satisfying a series of legal and fiduciary
requirements. Portfolio construction techniques, dating from the
Nobel-prize winning work of Markowitz in the 1950s, attempt to address
these often-contradictory constraints in a principled way, but the
resulting optimization problems are computationally intractable for
large-scale (many asset) portfolios. Beyond that, classical methods
are known to be statistically unstable and highly sensitive to
estimated input, making them difficult to apply successfully to noisy
financial data.

By recasting portfolio construction as a constrained (penalized)
regression problem, we pose a new class of techniques for portfolio
optimization based on high-dimensional statistical theory. Exploiting
this connection to existing theory allows us to show that our
techniques have superior properties across a wide range of scenarios.
Furthermore, we show that our techniques are computationally many
orders of magnitude faster than what currently appears in the
literature. Finally, we show, empirically and in simulation, that they
produce portfolios which are competitive with, and in many cases,
superior to those produced by existing methods.


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Data-Driven Reduced-Order Modeling of Turbulent Flows
Many geophysical and engineering flows are turbulent. The chaotic and inherently multi-scale nature of turbulent flows poses major theoretical and computational challenges to efforts aimed at understanding, predicting, or controlling such flows. Accurate calculations of turbulent flows require using computationally expensive Direct Numerical Simulations (DNS) or Large Eddy Simulations (LES). Finding the optimal design or devising online control strategies for turbulent systems often involve conducting numerous DNS or LES runs, which can be formidable for many problems, in spite of the ever-growing computational power.

As a result of these challenges, recent years have seen a significant interest in developing models that have lower computational complexities but retain the key dynamics and essential features of the turbulent flows. These so-called Reduced Order Models (ROMs), if accurate, can be readily used for optimal design and real-time prediction/control of turbulent flows because they are computationally tractable. Development of accurate, predictive ROMs for turbulent flows have been actively pursued in the academia and industry; however, currently no robust, effective, generally-applicable framework is available to calculate accurate ROMs for fully-turbulent systems.

Of particular interest is finding a robust and accurate data-driven framework for calculating predictive ROMs. A data-driven approach is desirable because it can be generally-applicable and cost-effective, and does not require a full understanding of the underlying physical processes, which are not well understood for turbulent flows. Statistical and stochastic modelling approaches based on Proper Orthogonal Decomposition (POD), Dynamic Mode Decomposition (DMD), Linear Inverse Modeling (LIM), and Fluctuation-Dissipation Theorem (FDT) have been extensively applied to find ROMs of turbulent flows. Ideas from big data and machine learning have also emerged in the past few years.

Currently, none of these data-driven frameworks can produce ROMs for fully-developed turbulent flows with the accuracy and robustness that is needed for real-world engineering problems. A major challenge is that the reason(s) behind these inaccuracies and failures are unknown and in general, there is a wide gap between the progress in dimension-reduction techniques for turbulent flows at the theoretical level and the applications of these techniques to fully-turbulent, complex systems. The purpose of this study is to bridge this gap.

This study consists of three steps: 1) A predictive ROM for a fully-turbulent Rayleigh-Benard convection system, which is a reasonable prototype for many geophysical and engineering flows, is calculated using a novel method that is accurate but numerically expensive and not data-driven; 2) ROMs using the data-driven methods such as DMD and FDT are calculated and then tested and compared and contrasted with the accurate ROM calculated in Step 1 to identify the sources of their (potential) shortcomings and find possible remedies; 3) The most effective data-driven method found in Step 2 is applied to data obtained from numerical simulations and/or networks of sensors to calculate predictive ROMs for turbulent flows that are of industrial interest. In this poster, we introduce the novel method that is used in Step 1 and present preliminary results from Steps 1 and 2 to show the promises of this approach.

Speakers
avatar for Pedram Hassanzadeh

Pedram Hassanzadeh

Faculty, Rice University


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Deep Learning Approaches to Structured Signal Recovery
The promise of compressive sensing (CS) has been offset by two significant challenges. First,
real-world data is not exactly sparse in a fixed basis. Second, current high-performance recovery
algorithms are slow to converge, which limits CS to either non-real-time applications or scenarios
where massive back-end computing is available. We attack both of these challenges head-on by developing new signal recovery frameworks that learn the inverse transformation from observation vectors to original signals using deep learning techniques. When trained on a set of representative signals, these frameworks learn both a representation for the signals (addressing challenge one) and an inverse map approximating a greedy or convex recovery algorithm addressing challenge two). According to simulation results, our deep network frameworks closely approximate the solution produced by state-of-the-art CS recovery algorithms yet are thousands of times faster
in run time. The trade-off for the ultrafast run time is a computationally intensive, off-line training
procedure typical to deep networks. However, the training needs to be completed only once, which
makes the approach attractive for a host of sparse recovery applications. These applications are but not limited to magnetic resonance imaging (MRI), computed tomography (CT), coded-aperture imaging, and seismic data collection in oil and gas industry.

Speakers

Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Estimation of Risk Measures under Parameter Uncertainty
Many industrial applications, including reservoir modeling and management, are subject to uncertainty due to the lack of knowledge of model parameters. Estimating uncertainty in the quantities of interest (such as oil production, water cut, pressure) of the systems subject to random variables (such as reservoir equations with uncertainties in the geological parameters) is a challenging task due to the large computational effort involved with simulations of numerical models. Particular difficulties arise when information is needed regarding the tail statistics of the random quantities of interest, such as in risk-averse oil production optimization.

We consider a class of risk measures that assign a value to the overall hazard associated with unfavorable outcomes of the random quantities of interest (e.g., large losses in profit). Common risk measures, such as absolute semideviation or conditional value-at-risk, are difficult to estimate accurately due to the sampling deficiencies. We propose an approach to estimation of these risk measures that relies on their analytic properties, and statistical concepts, such as importance sampling. Our approach aims to reduce the number of samples required to estimate risk measures accurately, thus leading to the reduction of number of costly simulations. The efficiency of our approach is demonstrated on the problem related to the steady flow in porous media.


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: GPUs as Workload-Managed Resources: Challenges and Opportunities for Applications in the Oil and Gas Industry
General Purpose Graphics Processing Units (GPGPUs) have made significant computational inroads in a number of industries. In the oil and gas industry, for example, sophisticated algorithms have allowed those with requirements for the processing of seismic data via Reverse Time Migration (RTM) to minimize significant I/O bottlenecks through extremely innovative use of GPUs (e.g., Liu et al., Computers & Geosciences 59 (2013) 17–23). As GPU hardware improves, so does the toolchain that programmatically exposes these accelerators for numerically intensive computations in the oil and gas industry. Buoyed by successes involving RTM, and other applications within the industry, it is evident that organizations seek opportunities to further exploit GPUs. In addition to exposing architectural specifics like accelerators-per-GPU-socket and CUDA-core-count, there is a clear interest in regarding GPU memory as a bona fide resource. And owing to the simultaneous TOP500 plus Green500 appeal of high-density GPU configurations, there is clear interest in exposing aggregations of these powerful accelerators as resources that can be addressed by workload-management software - with an awareness that includes topological considerations like GPU memory, sockets, chassis and interconnect fabrics. Resource maps (RSMAPS) were added to Univa Grid Engine to better account for the depth and breadth of capabilities available from GPUs, and to account for their aggregated capabilities. Whereas RSMAPS are able to address numerous resource requirements involving GPUs, it is clear that this remains an area where additional capabilities need to be accounted for by the workload manager. By systematically reviewing increasingly sophisticated use cases for the uptake of GPUs as diverse and capable resources for computational HPC in the oil and gas industry, opportunities and challenges for abstracting these accelerators as workload-managed resources will be shared.

Speakers
avatar for Ian Lumb

Ian Lumb

Solutions Architect, Univa Corporation
As an HPC specialist, Ian Lumb has spent about two decades at the global intersection of IT and science. Ian received his B.Sc. from Montreal's McGill University, and then an M.Sc. from York University in Toronto. Although his undergraduate and graduate studies emphasized geophysics, Ian’s current interests include workload orchestration and container optimization for HPC to Big Data Analytics in clusters and clouds. Ian enjoys discussing... Read More →


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: HPC Simulation of Hydraulic Fracturing in Three Dimensions
We present a new high performance computing software solution, ARGOS, for three-dimensional, coupled multiphysics simulation of fractured rock mechanics, fluid dynamics, and proppant transport. The capabilities are illustrated by some case studies of hydraulic fracturing applications. In contrast to other simulators that make simplifications such as dimension reduction, isotropy, and structured grids, ARGOS makes general allowances for anistropy, unstructured grids, fracture networks, and arbitrary configurations of multiple wellbores, perforations, and fractures. The wellbore model includes deformable casing, and the fractures dynamically propagate and expand in aperture as the rock deforms. The dynamic simulation of these coupled models produces realistic behavior of the physical system, which cannot be attained with simpler simulators.

ARGOS is validated by its high accuracy in analytic tests, and it is also applicable to general settings where analytic solutions are not available. We show that single fractures in simple settings, without proppant transport, are relatively insensitive to increasing fidelity of the analysis, but proppant transport and interactions of multiple fractures exhibit complex behavior that cannot be captured with simpler models and solvers. Our results demonstrate the strong significance of effects such as anisotropic in situ stress fields, stress shadowing, near-wellbore friction, proppant bridging, and slurry rheology. Proppant transport has a significant impact on slurry viscosity and hence the pressure applied to the fracture surfaces, resulting in complex fluid-structure interaction. Simulations must represent these effects in order to accurately estimate stimulation, proppant distribution, and production.

Efficient parallel algorithms make it possible to simulate large problems with high fidelity, using fine mesh resolution to accurately represent input fracture networks and stress fields. For scalability, we use domain decomposition techniques to solve an implicit time integration of the Darcy fluid flow in fractures and wellbores, coupled with a fast explicit time integration of the solid mechanics model.

Investigations with a powerful HPC software tool such as ARGOS are essential in evaluating the effect of mesh resolution on accuracy, as well as the negative impact of simplifying model assumptions and weak couplings that are made in simpler tools. This insight can also guide the design of simpler, faster solvers that neglect certain aspects deemed insignificant, in some conditions, by high-fidelity experiments.


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Hybrid Parallel Multiscale Reservoir Simulator
To sufficiently leverage increasingly powerful computer hardware, software applications must efficiently utilize all available cores. Amdahl’s law states that the bottleneck to the speedup and parallel scalability of an application is the extent of its serial part. Therefore, software applications must include optimal algorithms that are parallelizable on all available fronts. The fact that the algorithms employed are not embarrassingly parallel also poses significant challenges to architecting effective parallel implementation.

The multiscale method used to accelerate a fluid flow simulation framework helps in addressing these challenges. This method solves the problem on a coarse level and then prolongs it to fine level using a set of basis functions. These prolongation operators map between the fine-grid geological properties of reservoir model and the coarse grid that is used for simulation. This method has natural possibilities in algorithm parallelization because the coarse grid provides independent subdomains for concurrent computations.

We then use domain decomposition and multithreading in this hybrid parallel multiscale reservoir simulator. Domain decomposition is used to divide the problem into smaller subproblems that can be solved across distributed memory systems by parallel processes. Multithreading is employed within each of these parallel processes to further divide work and execute parallel threads with shared memory. Message Passing Interface (MPI), a well-known API, is used to facilitate communication between these parallel processes. OpenMP, a shared memory parallelization technology, is used for managing parallel threads.

A hybrid parallel MPI/OpenMP paradigm was implemented in a multiscale reservoir simulator as a part of a commercially available simulator. Remarkable speedup was achieved for problems of various complexities and sizes using state-of-the-art cluster systems with hundreds of compute nodes.


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Keeping our Energy Future Hydrated
Recently world leaders have become increasingly concerned about the water-energy nexus, a concept that refers to the necessity of water in energy production and the consumption of energy in the extraction, purification, and delivery of water. The need for the inclusion of water constraints within energy planning is more important today than ever before. In the US, 40% of freshwater extracted annually is used in thermoelectric power plants. Hydraulic fracturing is also known to be a water intensive process for energy production. Cooling-water scarcity has a tremendous impact on various nuclear and fossil-fueled power plants in the southeastern United States during the warmer summer months when they are forced to reduce production. My objective is to understand the dynamic relationship between energy and water commodities. In order to understand the dynamic correlations between these nonstationary commodities, I propose to explore their co-movement using wavelet coherence analysis and other time-varying spectral representations. Investment decisions based on this insight will be made under the umbrella of portfolio investment theory to determine the optimal risk-return tradeoff in the two commodities. The key to this investment strategy, and a focus of my research, is optimally estimating and forecasting the inverse of a sparse covariance matrix of complex time-frequency interactions. Correctly identifying the dynamic interactions of water and energy commodities not only creates a vehicle to improve upon current investment decision planning within the United States, but could also impact decision processes within countries that have high water stress. My research speaks directly to this important global challenge of the next two decades by helping investment planners make better decisions on how to allocate water to maximize energy returns while preserving potable water sources.

Speakers

Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Mitigating Disasters with Deep Learning from Twitter?
By including credible data extracted from the Twitter social networking service, the study of earthquakes and tsunamis is being transformed into a Big Data Analytics problem. The challenge of establishing geophysically credible tweets is considered through a combination of Deep Learning and semantics (i.e., knowledge representation). Although there remains cause for optimism in augmenting traditional scientific data with that derived from social networking, ongoing research (e.g., http://credit.pvamu.edu/MCBDA2016/Slides/Day2_Lumb_MCBDA1_Twitter_Tsunami.pdf) is aimed at providing utility in practice. The motivation for success remains strong, as establishing a causal relationship between earthquakes and tsunamis remains problematical, and this in turn complicates any ability to deliver timely, accurate messaging that could prove life-critical. Finally, the applicability of this approach to other disaster scenarios (e.g., oil spills) is considered.

Speakers
avatar for Ian Lumb

Ian Lumb

Solutions Architect, Univa Corporation
As an HPC specialist, Ian Lumb has spent about two decades at the global intersection of IT and science. Ian received his B.Sc. from Montreal's McGill University, and then an M.Sc. from York University in Toronto. Although his undergraduate and graduate studies emphasized geophysics, Ian’s current interests include workload orchestration and container optimization for HPC to Big Data Analytics in clusters and clouds. Ian enjoys discussing... Read More →


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Multi-Level Iterative Reconstruction with Elastic Surface and Body Waves on Unstructured Tetrahedral Meshes
We model the elastic time-harmonic waves using
the continuous Galerkin (CG) finite element method on unstructured tetrahedral meshes
and study the elastic full-waveform inversion with surface and body waves.
We project the scaled gradients onto stable subspaces
and further a multi-level scheme to stabilize our iterative reconstruction.
The procedure is illustrated via several computational experiments.


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Parallelization for Coupled Fluid-Heat Flow and Poroelastoplastic Geomechanics Simulation
In coupled method for flow and geomechanics simulation, the computational cost becomes very large as gridblock number increases. The flow problem and the geomechanics problem are separately solved and then iteratively coupled so that the interaction between flow in porous media and the rock deformation can be captured. In order to obtain fast simulation, in this work, a parallel solver is implemented in the coupled simulation to speedup solving of matrices. Parallel array assembly is also introduced so that array values can be efficiently assigned in very large linear systems.
The parallelization is based on a serial sequentially coupled non-isothermal fluid flow and elastoplastic geomechanics simulator. MPI-based parallel array assembly and the parallel solver PETSc are first implemented to the flow and geomechanics problems separately. Specifically, the parallel array assembly is for the sparse format of the Jacobian matrices, while the PETSc solver is introduced to replace the original serial solver. The solver is GMRES based on block Jacobi preconditioner. Then the two parts are sequentially coupled together. Different matrix decomposition methods are tested to understand its impact on solver time and solver iteration numbers. The reservoir model in this work has one million cells.
Results indicate that the parallel solver improves the linear system solving efficiency in both flow and geomechanics problems. Scalability is also observed in the parallel array assembly in both problems. In general, the speedup is better achieved in the geomechanics problem than in the flow problem. After the two parallelized problems are sequentially coupled together, the overall scalability of the coupled model is still honored. The optimum overall speedup for the coupled method reaches 14. It is observed that plasticity computation in the geomechanics problem results in load imbalance, as plasticity only occurs at certain gridblocks where the rock failure yield condition is triggered. Also, different matrix decomposition methods lead to different matrix solving time and different solver iteration numbers to convergence. This work can be straightforwardly applied to serial sequentially coupled models considering various fluid flow and rock deformation physics to obtain practical speedup in MPI-based environment.


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Pyrolytic Treatment of Soils Remediates Heavy Hydrocarbons and Enhances Fertility
Pyrolysis of biomass to produce biochar offers potential to improve soil agricultural quality and sequester carbon. Our research has integrated techniques frequently used in biochar production and soil remediation by thermal desorption to quickly remediate soils contaminated with recalcitrant heavy hydrocarbons from weathered oil spills. This approach preserves a fraction of soil organic carbon that are lost in other thermal technologies such as incineration.
We built a 0.5L fixed bed reactor to conduct bench-scale pyrolysis experiments, and determined appropriate processing conditions via thermogravimetric analysis. Plant toxicity studies were conducted using treated and untreated soils. Elemental analysis and SEM/FTIR microscopy are among the methods used to characterize pyrolyzed soils. Results show that pyrolysis is an effective way to reduce TPH below regulatory levels without significant formation of PAHs. In addition, plant trials have shown higher biomass production in pyrolyzed soils over contaminated and incinerated soils. Characterization experiments are ongoing to describe the effect that pyrolysis has on remaining heavy hydrocarbons, soil density and mobility, and agricultural properties such as carbon content, water holding capacity, and cation exchange capacity.
Results suggest that pyrolysis may have an important niche as a remediation strategy to quickly remove TPH below regulatory standards while preserving soil fertility and sequestering carbon.


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Robust Computation of Seismic Normal Modes using Rayleigh-Ritz Galerkin Method in a Spherically Symmetric Earth
The most popular software package Mineos is widely used in computing synthetic seismograms in a spherically symmetric non-rotating Earth. However, it fails to handle the catastrophic eigenvalue clustering. The eigenfunctions of different modes with very close eigenvalues are not orthogonal. On the contrast, our package can get the pure eigenfunctions verified by the classification theory.

Instead of introducing the minor vectors and shooting method used by Mineos, we choose the generalized Rayleigh-Ritz-Galerkin finite element method based on Buland's approach, which leads to a generalized eigenvalue problem. Buland's method works perfectly fine for toroidal modes, and however, it suffers from the ``extraneous" eigenfrequencies throughout the eigenspectrum. The eigenfunctions of these non-seismic modes have energies concentrated in the fluid outer core, which indicates that we need a novel technique to deal with the fluid region. Our method creatively projects out these non-seismic outer core modes by introducing the pressure besides the normal and tangential displacements. Based on Lehoucq's shift-invert strategy, we utilize a sparse, direct solver to speed up the algorithm.

Our package has a number of advantages over the finite difference and shooting method. The utilization of generalized Rayleigh-Ritz technique can handle the artificial sigularity at the center of the Earth. Therefore we can abandon the shooting method, which makes our approach more stable, mode independent, more efficient for the high accuracy requirement. Moreover, the Finite Element Method preserves the high accuracy across the boundary compared to Finite Difference Method. Most importantly, our successful projection in the fluid region can properly handle near degeneracy in eigenfrequency. We are also working on a parallel package to compute the non-symmetric three-dimensional seismic normal modes.


Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Sensor Technologies for Energy E&P and Environmental Monitoring
Rice Integrated Systems and Circuits (RISC) laboratory directed by Prof. Aydin Babakhani develops advanced sensors technologies for energy exploration and production. The RISC laboratory focuses on the fundamental sciences related to physics and electronics of silicon-based integrated sensors and antennas and takes the technology to oil fields after completing the research cycle. One of the focuses of the RISC laboratory is to eliminate the "valley of death" in deployment of university research.

The sensor research in the RISC laboratory covers two broad areas:

(1) permanent online monitoring sensors, and (2) miniaturized proppant-sized battery-less sensors with energy harvesting capability.

In the first category, the RISC laboratory has developed the world's first HPHT asphaltenes and corrosion sensor technology that can monitor paramagnetic chemicals such as asphaltenes in real-time. This is done all electronically and without using any chemicals or other consumables. The asphaltenes sensor technology was recently tested in a major oil field in Canada by one of largest Canadian energy producers. The second field trial was successfully completed in Permian Texas by one of the largest U.S. independent producers. This technology helps energy producers to significantly reduce the operating cost and minimize the environmental impacts by reducing the usage of chemical inhibitors for asphaltenes and corrosion. Another technology that is being developed in the first category is a sensitive THz gas spectrometer for on-line monitoring of H2S, CO2, and other polar molecular in surface facilities. In addition, the RISC laboratory is building a miniaturized mass spectrometer and a variety of microwave/radar sensors for online monitoring of chemicals in a multi-phase flow system.

In the second category, the RISC laboratory develops proppant-sized sensors than can be sent to reservoir during hydraulic fracturing jobs. These sensors aim to produce a high-resolution image of the fractures in the reservoir with spatial resolution of 1ft. They are also able to measure reservoir properties such as temperature, pressure, chemicals, stress, pH, etc. In addition to fracture mapping, the second category includes cement embedded sensors that are used to continuously monitor the quality of cement and report leakage of environmentally hazardous gases such as methane and CO2. These sensors can also be used to detect corrosion and water breakthrough in the cement.

Speakers

Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Studying the Lithosphere–Asthenosphere Boundary beneath the Western United States by High-Resolution Seismic Filtering based on Non-Convex Regularization of Radon Transform
The lithosphere-asthenosphere boundary (LAB) separates the rigid, cold lithosphere characterized by a conductive thermal regime from the underlying hot, convecting mantle asthenosphere. The LAB is also a rheological discontinuity that marks differential motion between tectonic plates and the underlying mantle, hence, studying this boundary will help us to understand plate motions, tectonics, and mantle convection. Ps and Sp phases that convert at the LAB are powerful tools to estimate the depth and the velocity gradient of the LAB, however, it is always been a challenge to study the LAB by Ps receiver functions due to interference of the primary conversions with strong crustal reverberations. In this study, we proposed a novel method to filter the crustal reverberations and to enhance the LAB signals based on non-convex regularization of Radon Transform. The Radon transfer sums signals exhibiting linear, parabolic or hyperbolic moveout in the original domain (t-Δ domain) to a single event in the new domain (Radon domain). Consequently in the Radon domain, the target signals can be isolated easily from the noise. To further enhance the spatial resolution in the Radon domain and decrease the computational cost, a pre-processing method was developed in our group based on the linear random transform in the frequency domain implemented with compressive sensing theory. The compressive sensing approach helps in recovering the sparsest solutions in the radon domain to underdetermined inverse problems. Instead of least square minimization, a non-convex minimization (Lp regularization with 0<p<1) is used to regularize the inverse problem. Our results show that at most only three iterations are needed for the non-convex minimization algorithm to reach the desired level of sparsity. Since non-convex regularized Radon Transform offers a benefit for signal isolation in Radon domain, this method can be a powerful tool to remove crustal multiples from Ps receiver functions and enable us to obtain better LAB images. Here, we show some results by applying this method to Ps receiver functions from broad-band seismic stations in USArray. For both synthetic and USArray receiver functions, the results clearly demonstrate that non-convex regularized Radon Transform can provide a superior reconstructed LAB image by filtering out the signals from crustal reverberations. We further propose to apply this method to reconstruct the common conversion point (CCP) stacked Ps and Sp receiver function images of the LAB beneath the western United States.

Speakers

Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC

4:00pm

Poster: Sunflow: Efficient Optical Circuit Scheduling for Coflows
Optical Circuit Switches (OCS) are increasingly used in cluster networks due to their data rate, energy and longevity advantages over electrical packet switches. Concurrently, an emerging crucial requirement for modern data-parallel clusters is to achieve high application-level communication efficiency when servicing structured traffic flows (a.k.a. Coflows) from distributed data processing applications. This paper presents the first OCS scheduling algorithm called Sunflow that addresses this requirement.

Preemption decisions are the key to any OCS scheduling algorithm. Sunflow makes preemption decisions at two levels. First, at the intra-Coflow level, Sunflow does not allow subflows within a Coflow to preempt each other. We prove that the performance of this strategy is within a factor-of-two to the optimal. We further demonstrate that under realistic traffic, performance of Sunflow is on average within 1.03x to optimal. Second, at the inter-Coflow level, Sunflow provides a framework for flexible preemption policies to support diverse usage scenarios. In the specific case of the shortest-Coflow-first policy, we find that Coflows on average finish just as fast in a Sunflow-scheduled optical circuit switched network as in a comparable packet switched network employing the state-of-the-art Coflow scheduler.

Speakers

Thursday March 16, 2017 4:00pm - 5:30pm
Exhibit Hall BRC