Tracks and Sessions

The Developer Congress agenda will cover the following tracks. You can view or download the sessions schedule here.

GPU Programming
OpenACC Programming
  • Part 1 – Intro to OpenACC – lecture + hands-on
    Instructor-led class that includes interactive lectures, hands-on exercises, and labs.
  • Part 2- Profiling and parallelizing with OpenACC
    Continuation of part 1 where you will start to profile and paralleize with OpenACC.
  • Part 3 – Expressing data locality and optimizations with OpenACC – lecture + hands-on
CUDA Programming
  • Part 1 – Intro to CUDA C/C++ – basic programming syntax – lecture + hands-on exercises. Learn the fundamentals of parallel computing with the GPU and the CUDA programming environment! In this class, you’ll learn about basic programming syntax through lecture format and practice with hands on exercises.
  • Part 2 – CUDA C/C++ basic optimizations, CUDA libraries – lecture + hands-on. Building upon Part 1, we will continue with basic optimizations and explore the CUDA libraries through hands on exercises.
Analysis Driven Optimization

An iterative sequence of writing a naïve matrix transpose, then using the profiler to guide optimization efforts, in CUDA C/C++

 

Presenter

Robert Crovella – NVIDIA

Presenter

Rohit Taneja – NVIDIA

Machine Learning / Deep Learning
Introduction to Deep Learning

An overview of Deep Learning – what is it, how is it similar to and different from ML, why GPUs are good at DL, and some practical application case examples.

Presenter


Robert Crovella – NVIDIA

How to Deploy Deep Learning in the Real World with Power Ecosystem

We will focus on the end to end data flow which is required when deploying deep learning into the world. Data flows from edge-of-cloud servers and mobile devices, back into enterprise data warehouses for periodic re-training. Updated neural networks and weights are then periodically broadcast out to inference servers and devices. This flow creates systems which are continuously learning and improving.

We will go into details of enterprise software tools and data center architecture. We will highlight differences between training and inference HW, as well as the process of optimizing a trained neural network for deployment on mobile devices.

 

Presenter

Sumit Sanyal – Minds.ai

Sumit Sanyal is the founder and CEO of minds.ai. He has 20 years of industry experience architecting integrated hardware and software systems to efficiently execute state of art algorithms. minds.ai is a leading provider of deep learning technology to global enterprises.

Nimbix Tutorial

Nimbix is an HPC & Cloud Supercomputing platform enabling engineers, scientists & developers, to build, compute, analyze, and scale simulations in the cloud. In this session we’ll give you an overview of Nimbix and walk through the practical steps to get started.

An Introduction to PowerAI

Slash the complexity of building your Deep Learning frameworks with innovations from IBM PowerAI.  In this session, you’ll be introduced to PowerAI and the motivation behind it. We’ll demonstrate an install and talk about what’s coming next.  Bring your laptop and we’ll take a tour of Nimbix, walk you through a framework activation and examples so you’ll have a head start for our deep dive sessions!

Presenter

Franck Barillaud – IBM

For the last 10 years, Franck Barillaud has been the WW Lead Cloud Architect for the ISV organization in IBM Systems Business Unit. He’s responsible for designing and supporting the Power Development Platform (PDP), a Power Cloud infrastructure for ISVs and BPs, which hosts over 5,000 AIX, IBM i and Linux on Power engagements from over 600 ISVs each year enabling solutions on Power that drive hardware sales. He’s also leading a cross IBM team to define, create and operate Power Based Cloud infrastructure in support of key projects for the transformation of IBM Power systems.He is the lead architect for the Power Innovation Cloud (PIC) to incubate innovative solutions around Power 8 accelerators (NVIDIA GPUs, CAPI/Flash Systems, FPGAs).
Franck own 32 patents and has been the recipient of multiple innovation awards.

Presenter

 

Jason Furmanek – IBM

Jason Furmanek has been a developer in IBM Power System for 15 years in various capacities including development on OpenGL graphics libraries, video encoding solutions on IBM’s Cell Processor, systems software, and open source community tools for Linux and Power Systems. He was the lead for Linux on Power in the Lab Services team for part of that time as well. Jason, Chief Engineer on IBM’s PowerAI product, is a big proponent of open systems, open software and open conversation.

Introduction to Tensorflow

TensorFlow is a very popular open-source library for Machine Learning. In this workshop you will get an introduction to Tensorflow, as well as get to try it out yourself through four exercises. In the end, you will have built a machine learning system that uses a deep-learning neural network to classify digits written by humans. The dataset used is called MNIST and is considered the Hello World of machine learning.

Presenter

Magnus Hyttsten – Google

Magnus Hyttsten has more than 20 years experience working on distributed computing systems, and is currently working as a Developer Advocate in the TensorFlow team at Google. In his spare time, he enjoys hiking, and fixing error messages generated by statically typed languages.

Inference System for Deep Learning

This session is about the leading technologies in industry of Deep Learning (DL) inference system. It includes interactive lectures and hands-on exercises. You will learn what an inference system for DL is and learn the key factors in designing a high performing inference system. You will learn what the PowerAI Inference Engine is, and how to generate a new DL FPGA accelerator from a DNN model without any coding requirements. With the real hardware environment on cloud, you can test drive the performance advantage from FPGA and CAPI for learning acceleration deep learning. We will also explain why CAPI/OpenCAPI will be important for deep learning acceleration.

Presenter

Dr. Chao Zhu – IBM

Dr. Chao Zhu is a research staff member from IBM China Research Lab. His research area includes cloud computing technology, FPGA acceleration for deep learning, containers and high speed network processing. He is the chief architect of SuperVessel cloud and the open source maintainer of DPDK on POWER. He has filed more than 10 patents.

Presenter

Kenneth Hill – IBM

Kenneth Hill received his B.S. and M.S. degrees in Computer Engineering in 2011, and Electrical and Computer Engineering in 2012 from the University of Florida, Gainesville. He is pursuing his Ph.D. in Electrical and Computer Engineering from the University of Florida with the NSF Center for High-Performance Reconfigurable Computing (CHREC). Kenneth is currently exploring the challenges of applying reconfigurable architectures for problems in the computational sciences. His research interests include: reconfigurable computing, heterogeneous systems, high-level synthesis tools for parallel computing, and parallel computing architectures.

Chainer Deep Dive

Preferred Networks applies cutting-edge deep learning technology to applications. In this session, you’ll be introduced to Chainer, a flexible framework for neural networks and get hands-on experience with Chainer by running a Jupyter notebook. Increase your expertise on the Chainer framework and experience fast, scalable, easy machine learning with OpenPOWER, GPUs and Docker on the NIMBIX public cloud. Bring your laptop.

 

Presenter

Nobuyuki Ota – Preferred Networks America

Caffe Deep Dive

Increase your expertise on the Caffe framework through hands-on guided exercises. You will experience fast, scalable, easy machine learning with openPOWER, GPUs and Docker on the Nimbix public cloud. You’ll look at image classification with a trained model. Train a model of a small dataset with DIGITS and image classify with your own trained model. Compare different techniques and explore the ”Model Zoo”.

Presenter

Clarisse Taaffe-Hedglin – IBM

Clarisse Taaffe-Hedglin is a High Performance Computing and High Performance Data Analytics (HPC/HPDA) executive architect, responsible for performance analysis and benchmarking on IBM Power Systems. She works in the IBM Systems Worldwide Clients Centers, helping clients define their HPC/HPDA strategy. Clarisse’s background is in numerical analysis and parallel systems optimization.

Presenter

Franck Barillaud – IBM

For the last 10 years, Franck Barillaud has been the WW Lead Cloud Architect for the ISV organization in IBM Systems Business Unit. He’s responsible for designing and supporting the Power Development Platform (PDP), a Power Cloud infrastructure for ISVs and BPs, which hosts over 5,000 AIX, IBM i and Linux on Power engagements from over 600 ISVs each year enabling solutions on Power that drive hardware sales. He’s also leading a cross IBM team to define, create and operate Power Based Cloud infrastructure in support of key projects for the transformation of IBM Power systems.He is the lead architect for the Power Innovation Cloud (PIC) to incubate innovative solutions around Power 8 accelerators (NVIDIA GPUs, CAPI/Flash Systems, FPGAs).
Franck own 32 patents and has been the recipient of multiple innovation awards.

Play with Deep Learning Models

Come to this session for another opportunity to see the deep learning frameworks in action, you’ll have access to a Nimbix instance and trained models you’ve worked on in the framework sessions and AI Vision session for an opportunity to run on your own and explore new ideas. Experts from the ML/DL track will be on hand for additional guidance. Bring your own latop to expand on what you learned!

Presenter

Clarisse Taaffe-Hedglin – IBM

Clarisse Taaffe-Hedglin is a High Performance Computing and High Performance Data Analytics (HPC/HPDA) executive architect, responsible for performance analysis and benchmarking on IBM Power Systems. She works in the IBM Systems Worldwide Clients Centers, helping clients define their HPC/HPDA strategy. Clarisse’s background is in numerical analysis and parallel systems optimization.

Image Analytics with AI Vision

This is an instructor-led class that includes interactive lectures, hands-on exercises, and labs. You’ll learn what AI Vision is and how it could help you to develop a customized DL model for your images or video streams. You could learn how AI Vision could support the entire DL development cycle, including the data labeling, data preprocessing, model training, and inference API deployment. You will practice how to use AI Vision to train a model with image files, and you could use the new deployed API to test the image recognition from the new models. On the other hand, we will introduce the framework with which deep learning developer could inject their own DNN design and algorithms into AI Vision. DL Insight is a tool which could work with AI Vision and help developer monitor and optimize their DNN performance and accuracy easily. In this session, deep learning developer could also learn how to use DL Insight to optimize the accuracy of their deep neural network.

 

Presenter


Dr. Yubo Li
– IBM

Dr. Yubo Li is a Research Staff Member at IBM Research, China. He is the architect of the GPU acceleration and deep learning service on SuperVessel, an open-access cloud running OpenStack on OpenPOWER machines, and the AI Vision (VisionBrain), a deep learning platform for customized visual recognition. He is responsible for deep learning infrastructures design, and GPU enablement and optimization on cloud.

Develop-Port-Optimize on Power Systems and GPU
Performance Analysis and Optimizations

Performance analysis, tuning and optimization of applications and its system environment have become increasingly important to exploit today’s compute environments. In this session, we will review Amdahl’s law to evaluate acceleration with GPUs to illustrate the challenges in making practical use of accelerators. Performance tools, including monitoring, tracing, profiling tools, etc., will be introduced for collecting performance data for analysis. We will cover performance analysis and tuning for processor performance, pipeline efficiency, system performance, and GPU performance. Come join our discussions on thread parallelization, IO throughput/latency, utilization of system resources and so on. Bring your performance problems and questions, the “Doctor” is IN!

Examples of Porting Applications to OpenMP4.5

This session focuses on examples to offload execution of kernels from CPUs (host) to GPUs using OpenMP4.5. We will start with managing data and memory using IBM’s implementation of OpenMP4.5 using “data map” directives and OpenMP4.5 API to allocate memory on devices (GPUs). Using a simple example based on daxpy-like operations , 4-5 different options for managing memory and data will be presented. We will review a few examples with complex data structures and applications using nested parallelism, and concurrent execution on CPUs and GPUs.

 

Presenter

Dr. Leopold Grinberg – IBM

Dr. Leopold Grinberg is a Research Staff Member in IBM’s Research Division. Leopold has vast experience in high performance and parallel computing. Dr. Grinberg is a technical lead of two Centers of Excellence established between the Department of Energy’s National Laboratories, IBM and NVIDIA. He oversees development of applications for IBM’s advanced computing architectures (IBM’s hybrid CPU+GPU systems). Dr. Grinberg focuses on algorithm development and performance optimization for massively parallel solvers on hybrid CPU+GPU platforms. He has expertise in CFD, Spectral Element, Particle Dynamics, seismic, transport solvers; C/C++, Fortran, CUDA, MPI, OpenMP4.

Lessons Learned from Benchmark Optimizations

In March 2015, IBM published results for a computational finance benchmark on a POWER8 system. The performance of this multithreaded C++ application exceeded all then current traditional systems, and one of our records has yet to be broken. In this talk we will discuss several factors that led to this result, including the importance of the right compiler, compiler options and libraries. We will also discuss how we leveraged knowledge of the POWER8 architecture – cache hierarchies, memory bandwidth, SMT and NUMA effects -to guide us to the best solutions for a given benchmark problem. If ultimate performance is the end goal then sometimes you just have to try things, and we will discuss the kinds of things we tried and what we found. Although the benchmark in question was an HPC benchmark, we will try to relate the findings to the broader range of C/C++ applications.

Nimbix Tutorial

Nimbix is an HPC & Cloud Supercomputing platform enabling engineers, scientists & developers, to build, compute, analyze, and scale simulations in the cloud. In this session we’ll give you an overview of Nimbix and walk through the practical steps to get started.

Bring Your Laptop and Let’s Power Optimize Your Code

POWER8 hardware is not your grandfather’s hardware! No more big endian challenges, all the major Linux distributions just run on POWER8. So, porting applications to Linux on POWER is simply a recompile and go! now, right!? Yep – almost. There are a few places where there are some minor differences from Lintel, but we’ve got you covered. With this session we will show you how to port applications to Linux on POWER, introduce you to some of the brand new development resources for LoP, give you insights for building, testing, packaging and getting support for applications. We’ll also go into the basics of tuning your applications for POWER and point out some of the common resources to making porting and development simpler. We’ll also introduce you to the basics of the common tools for optimizing applications to take advantage of some of the POWER8 hardware features.

Presenter

Ganesan Narayanasamy – IBM

Ganesan Narayanasamy is OpenPOWER ECO system leader for Education and Research , leading the Global OpenPOWER Academic Initiative. Ganesan collaborates with educational institutions world wide to help faculty and students to build OpenPOWER based solutions including HPC , Deep Learning and Cloud leveraging IBM resources and platforms available for academia and research organization . Ganesan has been with IBM since 1996 and has over 20+ years of industry experience across various aspects of the business, including product development , product management, HPC Benchmarking and Solutions.

Presenter

Gerrit Huizenga – IBM

Overview of Developing Applications on Power

There are a rich set of developer tools and resources available on Power. You’ll find that most of your favorite developer kits and environments are exactly the same on Power Linux as other Linux environments. This talk will review the tools available, how to locate resources, and will provide examples. We will then demonstrate via exercises how to use these facilities.

Analytics For Data Science using Apache Spark
Intro to Spark on Power

Want to learn more about how to accelerate your Hadoop workload 100x with Apache Spark? Industry experts will guide you to deploy and configure data analytic solutions (Spark, Hadoop and Zepplin) for POWER systems. Learn what is already out there and how to get access quickly and easily. We will have hands on activities to help you configure Apache Bigtop/Hortonworks Repo and see how to build a Hadoop/Spark Cluster using Docker/bare metal. Easily start services and become a voracious consumer to do more with your data using Spark and Zepplin.

Presenter

Amir Sanja – IBM

Amir Sanjar has many years of experience in big data software and solution development at companies including IBM and Canonical. He is the inventor of several patents in areas of enterprise solution automation and wireless/cell technology. Currently, he leads big data ecosystem and ISVs Power enablement at IBM

Accelerate Spark on Power

Now that you have the basics down for setting up and starting Spark services. Learn the tips and tricks specific to POWER for run-time optimization. Show off your new skills and get answers faster than anyone else. Bring your laptop and code you want to accelerate. We will go through an overview about the technologies and how IBM’s open sourced GPUEnabler and optimizations for Spark on POWER can accelerate your code. We will review Apache Spark tunables and its impact on workload execution. We will learn how to fine tune Spark workloads to take advantages of key hardware characteristics to optimize and accelerate code on OpenPOWER Systems.

 

Presenter

Raj Krishnamurthy – IBM

Raj Krishnamurthy designs and develops system stacks consisting of software and hardware elements for emerging and contemporary data analytics workloads. He has been a technical staff member in the Systems division at IBM since 2006. His work has impacted several platforms, software products, and roadmaps in IBM—both on mainframes and Power Systems. Raj holds 100+ patents and has written a number of external peer-reviewed publications. Raj holds a PhD in computer science and an MS/BS degree in electrical engineering.

DNA analysis with GATK using Spark and acceleration

No Bioinformatics expertise needed for this session.
This session will discuss a gene sequencing pipeline as a practical example of an end-to-end Spark-based solution. We first introduce the problem and the major stages of the pipeline, and then discuss GATK4, an open source Spark-based genome sequencing pipeline from the Broad Institute. We discuss how to install and run this pipeline on OpenPOWER servers and discuss some initial performance results. Next, we discuss a number of enhancements to the base configuration. FPGA-based data compression, CRAIL with Spark for enhanced shuffle performance, and FPGA and GPU acceleration for some of the more compute-intensive stages. We explain how the acceleration technologies integrate with Spark, and discuss the benefits these acceleration technologies can bring today and in the future.

Presenter

Zaid Al-Ars – TU Delft

Zaid Al-Ars is an associate professor at the Computer Engineering Lab of the TUDelft, where he leads the research activities of the big data research theme of the lab. His work focuses on developing HPC infrastructures to enable the efficient processing of big data applications using optimized solution alternatives for system performance, memory, power, etc. Dr. Al-Ars is also cofounder of Bluebee, a big data company specialized in high performance genomics solutions. Prior to joining the TUDelft, Dr. Al-Ars spent a number of years in the Product Engineering Group of Infineon Technologies and Siemens Semiconductors in Munich, Germany.

Presenter

Frank Liu – IBM

Spark Terasort and CRAIL

This session discusses the Terasort benchmark ( sortbenchmark.org ). It explains what is possible to achieve with OpenPOWER systems ( an OpenPOWER system is the current #1 ) by leveraging technologies like NVMe and 100Gb networking. We then discuss a Spark-based Terasort. We show what limits Spark performance, and introduce the open source CRAIL technology, integrated with Spark as a solution to overcome these dependencies. We demonstrate how a high-performance Spark/CRAIL Terasort can be built from open source on OpenPOWER.

High Performance Smarter Networks

Lecture and demo showing TF scalability on RDMA with POWER systems

Presenter

Scot Schultz – Mellanox

Open Platform for DBaaS on POWER
Open Platform for DBaaS on POWER Fundamentals

Come learn about OpenStack, Trove and Database as a Service. What hardware and software considerations should we make when deploying DBaaS on POWER systems? Experts will demo DBaaS with Open Source Databases including Redis, PostgreSQL, Mariadb. MySQL and MongoDB

OpenStack Private Cloud & DBaaS

Join us for a hands on exercise to access a DBaaS environment. We will configure a private cloud (compute/network/storage) and create a database instance. Demos for advanced configuration, security and clustering will be discussed and practiced. Bring your laptop and prepare to get your favorite open source database configured in minutes!

Database Programming in a DBaaS Environment

Development professionals can do more with DBaaS. We will review DevOps fundamentals and how Trove can help automate DBaaS tasks. Different types of databases will be evaluated , SQL or NoSQL, Mongo, etc. We will work with a sample dataset and prepare a MongoDB. Building on that, we will set up simple DB application. Questions on updates, upgrades and development vs production will be discussed.

 

Presenter

Marcelo Perazolo – IBM

Marcelo Perazolo is the lead architect for the IBM Cloud Toolkit for OpenPOWER and the main architect for Operational Management and DevOps for the IBM Systems Cloud Solutions team. He is located in RTP, NC, and received his MSEE and BSEE degrees from UNICAMP, then started his career at IBM in 1990, with more than 26 years of experience on Infrastructure and Platform Management solutions. He drives planning & strategy to exploit Open Software to build Converged and Hyperconverged infrastructure solutions. He is active in multiple organizations (such as OpenPOWER, OpenStack, OASIS and DMTF) and has dozens of patents and articles published. Marcelo has also promoted cooperation with many Academia and Industry partners in several opportunities and has previously acted as general chair, track chair, member of the technical committee or as featured speaker in several events. Marcelo’s main focus is on furthering adoption of OpenPOWER Systems in the marketplace.

 

Presenter

Tim O’Keefe – IBM
Tim O’Keefe is a Certified Human Factors Professional and lead designer in the IBM Power Systems Design group with more than 25 years of experience in user research, user interface design and human factors. He has extensive experience using IBM Design Thinking principles and working with customers to develop compelling and easy-to-use solutions. He is skilled in understanding and solving complex problems in creative ways with 10 issued and 6 filed patents.

Firmware and OpenBMC
Firmware and OpenBMC

Firmware Developers and System Operators Unite! Come to our Congress and develop code for a Base Management Controller (BMC). Meet, learn and network with today’s Titans of the Open BMC project. Bring your favorite laptop and editor for a hands on experience to write code and learn how a BMC operates. Through labs and tutorials, write your own customized common BMC management tasks (i.e. telemetry for environmental and power metrics). Help us create the first open Redfish design and verification suite. Leverage our experts to help you deploy and run Open BMC on your hardware.

Learn directly from the development experts how the On Chip Controller(OCC) interacts with the OpenBMC to manage system power, temperature, and performance. Modify OCC code to create new sensors, change system performance characteristics, or add new management capabilities.

  • Part 1 – Intro to OpenBMC and Firmware
  • Part 2 – OpenBMC Tutorial
  • Part 3 – On Chip Controller (OCC) Tutorial
  • Part 4 – Hackathon Plan
  • Part 5 – Let’s Code
  • Part 6 – Bring Your Own Board (BYOB)

 

Presenter

Adi Gangidia – Rackspace

I help build and scale the next generation, cost-effective openPOWER Barreleye / Zaius Servers (featuring IBM’s Power 8 / 9 processors) so that cloud computing gets more affordable. I do whatever it takes for Rackspace openPOWER initiatives to get into production. Ask me about why we are are excited about Power and open source firmware at Rackspace!

Presenter

Chris Austen – IBM

Chris Austen manages the Emerging Opportunities department inside POWER Systems Enablement. This role allows Chris and his team to develop and grow technologies such as the OpenBMC Project, SDN, and OpenStack as well a other lesser known works which improve the quality of IBM Systems. In 2015 Chris and a small team developed the OpenBMC for Rackspace’s Barreleye. A complete top to bottom open source POWER8 data center server. Since then the OpenBMC has grown in acceptance and supports multiple platforms. Chris joined IBM in 2000 and has to performed many roles in Firmware such as build, test, development, architecture, and its management. He developed the first IPMI stack for POWER servers and was the firmware lead for the POWER Blade line. More recently Chris helped develop the Monocle Patch management project, run the PowerVM social blogs aimed at educating, and oversaw the POWER8 overhaul of Host code resulting in the entire Host (BIOS) source code being available on github. Chris lives in Austin Texas where he a father to three and a husband to a newly minted Nurse RN.

Presenter

Chris Cain – IBM
Graduated from North Dakota State University with BS in Electrical & Electronics Engineering. Earned Master of Computer Engineering degree while working at IBM. Started at IBM in 1990 working in Local Area Network firmware development and testing (Token-Ring, Ethernet, FDDI, ATM) for the AS/400 / iSeries systems. Moved to the Power control team where I worked on firmware dealing with control and communication with power supplies and fans for the IBM servers. In 2006 I moved to the Thermal Management team where I worked on service processor and embedded firmware. This included design, development and testing of firmware to communicate with the TPMD card and OCC chip, as well as the firmware embedded on those devices. I live near Rochester, Minnesota with my wife and our four boys (ages 13 – 19).

Blockchain
Introduction to Hyperledger Fabric v1.0 running on Power

Business transactions and networks have been continuously evolving. The next revolution is here, industries are capitalizing by implementing Blockchains, improving transaction efficiencies and security. Join the IBM leaders who work with Hyperledger Fabric, explore Blockchain technology and come get your hands dirty! Instructors will guide you to setup your own blockchain network using Docker images and deploy sample chain code . Bring your laptop and let’s code.

 

Presenter


Srirama Sharma – IBM
I am Smita Bodepudi, working in IT industry over 18 years. Currently a developer in
Power Open Source Solutions team working on enablement of Blockchain Technology(Hyperledger Fabric) on IBM Power. Other role I have is Continuous Integration tech lead(working on both functional and performance) for other Open Source packages handled by my team namely Apache Hadoop/Bigtop etc.