The Past, Present, and Future of High Performance ... · The Past, Present, and Future of High Performance Computing in CADRE BJ Lougee Center for the Advancement of Data and Research

The Past, Present, and Future of High Performance Computing in CADRE

BJ Lougee

Center for the Advancement of Data and Research in Economics

Federal Reserve Bank of Kansas City

The views expressed in this presentation are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of Kansas City or the Federal Reserve System.

FRBKC’s Mission

● Fed has three main functions– Conduct the nation’s monetary policy

– Provide and maintain an effective and efficient payments system

– Supervise and regulate banking operations

FRBKC’s Mission

● Monetary policy– Why:

● We support a healthy economy that promotes stable growth and provides opportunities for all Americans.

– What: ● We participate in the nation’s monetary policy process to

promote maximum employment and price stability over the long run. These are goals given to us by Congress.

FRBKC’s Mission

– How: ● We produce mission-specific and thoughtful research and

analysis that informs the Bank president’s views and influences policymakers at all levels.

● We are a leader in providing computing resources and data warehousing to enhance research and analysis on macroeconomic, payments, banking, community development and regional issues

● We have regional branch offices with accessible, expert resources such as economists, examiners, community development specialists, economic educators and other specialists

● We connect Main Street views to regional and national policy through diverse boards of directors and advisory council members

Past: UNIX CLI

In the Beginning … Was the Command Line

- Neal Stephenson

Past: UNIX or Desktop

● Individual desktop computers (* Picture is not of actual work station)

✔ Easy to use

✔ No scaling

● UNIX server (Solaris)✔ Scales

✔ Not as easy to use as desktop computer

✔ Expensive

Past:Segmentation fault (core dumped)

● Moved to Linux Cluster of Workstations (COW)

Past:Segmentation fault (core dumped)

Past:Segmentation fault (core dumped)

● Move to Linux Cluster of Workstations (COW)✔ Less expensive

✔ Open source software

✔ Hardware scales at a better price point

✔ Can solve problems that can’t fit on desktop

✔ Not as easy to use as desktop computer

✔ Poor utilization

✔ Demolition Derby not optional

Past:Do you work here?

● Library’s Role– No real interaction

– Library wasn’t on anyones radar

Present:It takes a lot of flops to calculate the answer ‘42’

● Move from COW to HPC✔ Better resource

utilization

✔ Can scale beyond a single box

✔ Heterogeneous compute environment

✔ Need more training for efficient use

Present : Shiny New Toys

● More computational abilities added– General Purpose Graphic

Processors (Nvidia)

– Many Integrated Core architecture (Intel Xeon Phi)

● New software and computational paradigms– MapReduce (Spark)

– Machine Learning (TensorFlow)

– Reproducibility (Containerization)

Present:It takes a lot of flops to calculate the answer ‘42’

Present : Typical Types of Jobs

● Can be largely broken into two categories– High-throughput computing (HTC)

● Parameter sweeps

– High-performance computing (HPC)● Finding solution to system of non-linear equations

● Types of parallel architectures used– OpenMP

– MPI

– CUDA

● Most codes are embarrassingly parallel● Codes both cpu and memory bound

Present : Typical Types of Jobs

● Example 1– Heterogeneous agent models with incomplete asset

markets. Interested in how individuals choose to save and work in response to labor income risk

● Solve functional equations by using discrete approximation

● Uses CUDA with NVIDIA K40● Has started solving his functional equations by recasting

them to look like partial differential equations– Using finite difference methods to solve

Present : Typical Types of Jobs

● Example 2– Solve for optimal choice today give expectations

about the future● Solving system of non-linear equations on a gird● At each point in the grid each non-linear equation has to

hold● Can have many dimensions ● Uses Fortran with OpenMP to solve the guesses at each

point in the state space● Update initial guess and continue until converged

Present : Typical Types of Jobs

● Example 3– Natural Language Processing: textual analysis on

newspapers to understand media coverage of fiscal policy

● Papers from 1800s were scanned in using OCR● Uses Python to spell check each word in article● In parallel extract keywords from articles● Run regression

Present:Pass the Baton

● Library’s Role– Procurement of data

– Now we talk to each other but there is a hand off

Present / Transition: NSF Defined

● Traditional Physical Infrastructure– Facilities

– Instrumentation/ equipment

– Research Platforms● Vehicles● Multi-user

facilities

● New Cyberinfrastructure– Advanced computing

resources

– Digital libraries

– Shared data and information bases

– Research and education networks

– Distributed user facilities

– Standards and protocols

Present / Transition: Defining Cyberinfrastructure

● Cyberinfrastructure – Consists of computing systems, data storage

systems, advanced instruments and data repositories, visualization environments, and people, all linked by high speed networks to make possible scholarly innovation and discoveries not otherwise possible.

Present / Transition:People are Key

● Cyberinfrastructure personnel– Needed knowledgeable personnel to manage the

transition from COW to HPC and run new environment

– Training is essential for user base ● Front-line personnel ● Code optimization and debugging

Present / Transition: Collaboration

● Access to national resources – XSEDE

● Collaboration – Campus Champions

– ShowMeCI, KanShare

Future:Training

● CADRE– Training and user education

– Managing expectations

– Reproducible code and containerization

– Continue to find ways to make the transition of moving from local workstation to parallel computational resources easier

Future:Collaboration

● Ongoing Collaboration– Plan for the ending of XSEDE

– Expand partnerships with Cyberinfrastructure personnel within ShowMeCI and KanShare

– Grow the technological know-how for Friction Free Networking

– Public outreach

Future:It’s a group effort

● Library’s Role– Great team

cohesiveness

– Library, Researchers, and HPC work together

● Technology● Workflows● Preservation

Future:It’s a group effort

FIN

Attributions● HP-HP9000-C110-Workstation: Copyright 2015 by Thomas Schanz (https://commons.wikimedia.org/wiki/File:HP-HP9000-C110-

Workstation_10.jpg) reprinted courtesy of Creative Commons Attribution-Share Alike 3.0 Unported license ( https://creativecommons.org/licenses/by-sa/3.0/deed.en )

● Sun Enterprise 450: Copyright 2007 by Tommy Karlsson (https://commons.wikimedia.org/wiki/File:SUN_Ultra_Enterprise_450.JPG ) reprinted courtesy of licensed as public domain

● Tux: Copyright 1996 by Larry Ewing ( https://commons.wikimedia.org/wiki/File:Tux.png ) ; The copyright holder of this file allows anyone to use it for any purpose, provided that the copyright holder is properly attributed. Redistribution, derivative work, commercial use, and all other use is permitted. Attribution: [email protected] Larry Ewing and The GIMP

● Demolition Derby: Copyright 2007 by Bill Lowenburg ( https://commons.wikimedia.org/wiki/File:West_End_Fair_Gilbert_PA_Demolition_Derby.JPG ); reprinted courtesey of GNU Free Documentation License ( https://commons.wikimedia.org/wiki/Commons:GNU_Free_Documentation_License,_version_1.2 ).

● NVIDIA GPU : Copyright 2012 by GBPublic_PR ( https://www.flickr.com/photos/gbpublic/8178512552 ). Reprinted courtesy of Copyright Holder under a Creative Commons License CC BY 2.0 (https://creativecommons.org/licenses/by/2.0/)

● Intel Xeon Phi: Copyright 2016 by Portal GDA ( https://www.flickr.com/photos/135518748@N08/27729423232 ). Reprinted courtesy of Copyright Holder under a Creative Commons License CC BY-NC-SA 2.0 ( https://creativecommons.org/licenses/by-nc-sa/2.0/ )

● Apache Spark: Copyright 2018 ( https://commons.wikimedia.org/wiki/File:Apache_Spark_logo.svg ) reprinted courtesy of Apache License 2.0 ( https://www.apache.org/licenses/LICENSE-2.0.html )

● Tensorflow : Copyright 2017 ( https://commons.wikimedia.org/wiki/File:Tensorflow_logo.svg ) reprinted courtesy of Creative Commons Attribution-Share Alike 4.0 International Apache License 2.0 ( https://creativecommons.org/licenses/by-sa/4.0/deed.en )

● Singularity: Rights Singularityware ; BSD 3 ( https://opensource.org/licenses/BSD-3-Clause )