Seminar Information

4th AIST AI Workshop, "Workshop on Container Technologies for Convergence of HPC and AI/Big Data Analysis"

Attendance reception has closed.

All seminars in the Workshop are in English. No interpretation is provided.

At AIST, we are currently preparing to launch a cloud platform (AI Bridging Cloud, AIBC), with world-class AI processing performance, in 2018. In addition, at RWBC-OIL (AIST-Tokyo Tech Real World Big-Data Computation Open Innovation Laboratory), in order to implement deep learning, machine learning, and big data analytics efficiently and smoothly on ABCI and Tokyo Tech's TSUBAME supercomputer, we are working to create an open platform where HPC converges with AI and big data processing systems. In this workshop, seminars on the latest developments in HPC container technology and actual applications will be given by experts in each field, and we will share the latest trends and ongoing technical issues.

Title	4th AIST AI Workshop, "Workshop on Container Technologies for Convergence of HPC and AI/Big Data Analysis"
Date and time	July 10, 2017 (Mon) 13:30-17:30
Reception hours	12:50-17:00
Venue	(Please note that, from this occasion onward, the seminar venue has changed.) Asia Start-Up Office MONO, 14F Telecom Center Building East Hall, 2-5-10 Aomi, Koto-ku, Tokyo 135-0064
URL	https://mono.jpn.com/telecom-center-access/
Capacity	200 persons
Registration to attend	Closed
Admission	Free
Organizer	AIRC, AIST
Contact	AI Seminar Liaison
Seminar details	This seminar is an activity in “Development of Core Technology for Next-Generation AI and Robots” (Next-Generation AI Field),” a project commissioned by NEDO. This seminar is co-organized by AIST-Tokyo Tech Real World Big-Data Computation Open Innovation Laboratory (RWBC-OIL).
Points to Note If you become unable to attend, please cancel your registration promptly, out of consideration for others who would like to attend. Please register under your real name. AIST will not use the information you send for any purpose other than seminar management. No social event is scheduled after the seminar.

Program

13:30-13:35	Opening


13:35 - 14:00	Shinya Morino (NVIDIA) Title : GPU-computing with nvidia-docker in cluster environment.
	Outline NVIDIA is releasing nvidia-docker, and nvidia-docker is a tool to make GPU utilization easier under docker environment. In spite of nvidia-docker, we still need further consideration and integration with popular resource managers and schedulers to efficiently utilize GPUs in cluster environment. In this talk, challenges for effective utilization of GPUs in cluster environment will be discussed. Furthermore, topics for the next version of nvidia-docker will also be introduced.
	Profile Shinya Morino, Ph.D. Shinya Morino Joined NVIDIA in 2013, and has been working for education and customer supports on GPU-computing. Currently he is working as a DGX-1, whose software stack is strongly depending on docker.
14:00 - 14:25	Shuichi Ihara (DDN) Topic：Data resource and Security Management for Docker Container on Parallel filesystem
	Outline As container technologies such as Docker are used more frequently in HPC, AI, and big data systems, data resource management and data security are becoming increasingly important for large-scale storage systems and parallel file systems. To create a security system for the parallel file system Lustre, DDN has contributed to the Lustre community from an early stage. In this seminar, we explain the latest initiatives regarding data resource management and data security when using Docker Container on Lustre.
	Profile Shuichi Ihara, Senior Manager, Performance Engineering Lab, DDN After joining DDN in 2010, was involved with I/O benchmarks, performance tuning, and large-scale storage system proposals. Is currently involved in performance evaluation and proposals regarding next-generation storage systems and architectures, and is also in charge of Lustre development. Before joining DDN, worked at Sun Microsystems for 10 years. After gaining experience as a security consultant, was involved with HPC storage proposals, installation, and support as an HPC storage systems engineer, in Japan and other countries. From 2007, transferred to the Lustre development team, and worked on development of Sun HPC Software Stack, including Lustre, as a software development engineer.
14:25 - 15:25	Lucas Benedicic、Felipe Cruz（Swiss National Supercomputing Centre、Computational Scientist） Title: "Shifter: deploying Docker containers on HPC systems"
	Outline In this talk we will describe the experiences of building and deploying containers using Docker and Shifter on Piz Daint, the Cray XC50 supercomputer at CSCS. Docker is widely used by developers as a powerful tool for standardizing the packaging of applications across multiple environments, which greatly eases the porting efforts. On the other hand, Shifter provides a container runtime that has been specifically built to fit the needs of HPC. We will briefly introduce these tools while discussing the advantages of using them to fulfill the needs of specific workflows for HPC, e.g., security, high-performance, portability and parallel scalability. We will also present benchmarking tests that show the performance portability of certain workflows as well as performance results from the deployment of widely used non-trivial scientific applications. Finally, we will show how containerized application deployed with Shifter on Piz Daint can use specialized hardware like Cray Aries interconnect and NVIDIA Tesla P100 GPU accelerators.

15:25 - 15:40	break


15:40 - 16:40	Burak Yenier (UberCloud, CEO) Title:HPC-workloads in the Cloud and Container technologies

	Profile Burak Yenier is the CEO of UberCloud. He is an expert in large-scale, high availability systems and cloud. As an early SaaS proponent, Burak's management experience spans software development and operations. His most recent role was as the Vice President of Operations of a Silicon Valley SaaS company in banking. Burak built the company's cloud infrastructure and operations from scratch and for scale. He also managed all the data centers and the digital payment operations. Burak co-founded UberCloud in 2012. UberCloud has developed over 150 case-studies in every conceivable area of engineering simulation. The company publishes award-winning blueprints and best practices in cloud and technical computing. Burak simplifies the lives of engineers with powerful, easy to use compute environments in the Cloud. @BurakYenier https://www.linkedin.com/in/buraky/
16:40 - 17:05	Kento Aida (NII) Title: Application-Centric Overlay Cloud Utilizing Inter-Cloud
	Outline This talk presents a middleware development project to build big data analysis platforms on inter-cloud. Our middleware automatically creates virtual clouds customized for user applications by deploying Docker containers and configuring virtual networks.
	Profile Kento Aida received Dr. Eng. in electrical engineering from Waseda University in 1997. He became a research associate at Waseda University in 1992. He joined Tokyo Institute of Technology and became a research scientist at the Department of Mathematical and Computing Sciences in 1997, an assistant professor at the Department of Computational Intelligence and Systems Science in 1999, and an associate professor at the Department of Information Processing in 2003, respectively. He is now a professor at National Institute of Informatics from 2007.
17:05 - 17:25	Kuniyasu Suzaki (AIST) Title: Bare-Metal Container
	Outline Container technology is popular but has some problems. It does not allow to change the kernel. It is designed for multi-tenancy and makes difficult to measure power consumption for an application on a target machine. Bare-Metal Container (BMC) is proposed to solve these problems. BMC runs a container image on a remote physical machine directly with a suitable Linux kernel. It means that user can customize a Linux kernel for an application. BMC includes overhead to boot a Linux kernel, but the optimized application can surpass the overhead. It also measures power consumption and offers the information for low power execution. BMC allows to change a target machine easily and confirm the affinity between the application and the architecture.
	Profile Kuniyasu Suzaki is a senior researcher at ITRI (Information Technology Research Institute) of AIST (National Institute of Advanced Industrial Science and Technology). He got a bachelor and a master degree from Tokyo University of Agriculture and Technology and received his Ph.D. in Computer Science from the University of Tokyo. His research interests are computer security, operating system, and virtualization technology.
17:25 - 17:30	Closing