Accelerating Time to Science Using Cloud

P U B L I C S E C T O R
S U M M I T
Washington DC

© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.P U B L I C S E C TO R
S U M M I T
Accelerating Time to Science Using
Cloud
Sanjay Padhi
AWS Research Initiatives
Amazon Web Services
S e s s i o n I D 3 0 1 0 7 5
Manish Parashar
Office Director, OAC
National Science Foundation
Demián Arancibia
Astroinformatics Program
Chilean Ministry of Economy
Chris Wood
eBird Lead, Ornithology Lab
Cornell University

S U M M I T
Extraordinary Progress in Technology
• Two decades ago, microprocessors had 4 million
transistors
- Dual-core Itanium 2: 1.7 billion transistors
• Two decades ago, the internet had about a million
users
- Today, more than 1 billion
• Two decades ago about 15% of household had a
computer
- Today, nearly everyone owns a
computer in his pocket (smart phones) ENIAC (1943/1946) about 1,800 square feet
with 18,000 vacuum tubes, weighing ~50 tons

S U M M I T
National Academies Report

S U M M I T
Pillars for Modern Science:
Research Cyberinfrastructure

S U M M I T
AWS Global Infrastructure
21 Regions – 66 Availability Zones – 158 Edge Locations
11 Regional Edge Caches in 68 cities across 29 countries
Region & Number of Availability Zones
AWS GovCloud (US) Europe
US-East (3), US-West (3) Ireland (3)
US West Frankfurt (3)
Oregon (4) London (3)
Northern California (3) Paris (3), Stockholm (3)
Asia Pacific
US East Singapore (3)
N. Virginia (6), Ohio (3) Sydney (3), Tokyo (4),
Seoul (2), Mumbai (2)
Canada Osaka-Local(1)
Central (2) China
Beijing (2)
South America Ningxia (3)
São Paulo (3)
New Regions
Bahrain, Cape Town, Jakarta and Milan

S U M M I T
Evolution in Compute Services
Virtual Server, Container management, and Serverless Computing
Amazon EC2
Provides resizable cloud-based compute
capacity in the form of EC2 instances, which
are equivalent to virtual servers
AWS Lambda
Run code without thinking about servers.
Serverless compute for stateless code
execution in response to triggers
Amazon EC2 Container Service
A highly scalable, high performance
container management service

S U M M I T
Components
Automation and
orchestration
AWS Batch
AWS ParallelCluster
NICE EnginFrame
Storage
Amazon EBS
Amazon EFS
Amazon S3
Compute
Amazon
EC2 instances
(Compute and
accelerated)
Amazon EC2 Spot
AWS Auto Scaling
Visualization
NICE DCV
Amazon
AppStream 2.0
Networking
Enhanced
networking
Placement
groups
Elastic Fabric
Adapter
Amazon FSx
for Lustre
Simplified Provisioning: AWS Service Catalog
Service Catalog enables organizations to deploy and manage
AWS infrastructure and applications that reflect the
organization’s security and operational policies

S U M M I T
Amazon EMR
Amazon
DynamoDB
Amazon Kinesis
Data Analytics
AWS Marketplace
Amazon RDS
AWS Lambda AWS IoT Core
AWS
CloudFormation
Amazon Redshift
AWS
Service Catalog
Self-service based
AWS console
I Need a
Server
Broad Choices…
 Requires Security Policy
 Time consuming
 Incorrectly tagged
 Cost over runs
Amazon S3 Amazon EC2

S U M M I T
Example: How Notre Dame is using AWS Service Catalog

S U M M I T
Heterogeneity in Compute Resources
M5
General
purpose
Compute
optimized
C5
C4
Storage and IO
optimized
I3, H1
P3
Accelerated
computing
Memory
optimized
R4
D2
M4
X1/e
R3
P2
G3
F1
M5.24xlarge
• 96 vCPU,
• 384GB RAM
• Up to 25Gps n/w
• EBS only
• 9k EBS Mbps
• New Nitro light
hypervisor +
dedicated h/w
C5.18xlarge
• 72 vCPU,
• 144GB RAM
• EBS only
• 9k EBS Mbps
• Up to 25 Gbps
w/ENA
T2.2xlarge
• 8 vCPU,
• 32GB RAM
• EBS only
• 81 cpu
credit/hr
X1e.32xlarge
• 128 vCPU,
• 4TB RAM
• 2 x 1.9TB SSD
• 14k EBS Mbps
R4.16xlarge
• 64 vCPU,
• 488GB RAM
• SSD EBS
• 25 Gbps
H1.16xlarge
• 64 vCPU
• 256GB RAM
• 8 x 2TB HDD
• 25 Gbps
I3.16xlarge
• 64 vCPU
• 488GB RAM
• 8 x 2TB NVMe SDD
• 25 Gbps
I3.metal
• 36 cores/72
• 512GB RAM
• 8 x 2TB NVMe SDD
• 25 Gbps
D2.8xlarge
• 36 vCPU
• 256GB RAM
• 8 x 2TB HDD
• 25 Gbps
P3.16xlarge
• 8 GPU Tesla V100
• 5k CUDA/640 Tensor cores
• 488GB RAM
• 64GB GPU RAM
• NVLink p-2-p

S U M M I T
• 10s-100s of processing cores
• Pre-defined instruction set & datapath
widths
• Optimized for general-purpose
computing
CPU
CPUs vs GPUs vs FPGA for Compute
• 1,000s of processing cores
• Pre-defined instruction set and
data path widths
• Highly effective at parallel
execution
GPU
• Millions of programmable digital logic cells
• No predefined instruction set or datapath
widths
• Hardware timed execution
FPGA
DRAM
Control
ALU
ALU
Cache
DRAM
ALU
ALU
Control
ALU
ALU
Cache
DRAM
ALU
ALU
Control
ALU
ALU
Cache
DRAM
ALU
ALU
Control
ALU
ALU
Cache
DRAM
ALU
ALU

S U M M I T
Processing Capabilities: Accelerating time to results
Intel® Xeon® Scalable processor-based Amazon EC2 instances provide unique advantages to
accelerate HPC workloads
 Intel® Xeon® Scalable processors:
o Up to 28 cores delivering enhanced per core performance, and significant increases in memory bandwidth (6
memory channels) and I/O bandwidth and throughput (48 PCIe lanes)
o Intel® Advanced Vector Instructions (Intel® AVX-512): Intel® AVX-512 can handle your most demanding
computational tasks and accelerate performance for workloads
o Intel® AVX 512 delivers up to 2X more FLOPs/clock-cycle for HPC, analytics, cryptography and data compression
workloads5.
 Intel Software development Tools
o Allow users to take advantage of the Intel Xeon Scalable processor and quickly deploy a fully elastic HPC cluster
on AWS Cloud. Once created, the cluster provisions standard HPC tools such as schedulers, Message Passing
Interface (MPI) environment, and shared storage.

S U M M I T
Innovations in Infrastructure
Massively scalable performance
• C5n Instances will offer up to 100 Gbps of network
bandwidth
• Significant improvements in maximum bandwidth,
packet per seconds, and packets processing
• Custom designed Nitro network cards
• Purpose-built to run network bound workloads
including distributed cluster and database workloads,
HPC, real-time communications and video streaming
HPC stack on AWS
3D graphics virtual workstation
License managers and cluster
head nodes with job schedulers
Cloud-based, auto-scaling HPC clusters
Shared file storage Storage cache
Intel Xeon Scalable
(Skylake) processor
Featuring

S U M M I T
Security Requirements https://aws.amazon.com/compliance/

S U M M I T
Case Studies: Scientific Computing

S U M M I T
Elasticity in Computing: On Demand Auto-expansion to AWS
~60,000 slots using AWS spot instances. A factor of 5 larger than Fermilab capacity!
https://aws.amazon.com/blogs/aws/experiment-that-discovered-the-higgs-boson-uses-aws-to-probe-nature/

S U M M I T
Available in AWS Marketplace: Elastically bursting to AWS
https://research.cs.wisc.edu/htcondor/manual/v8.7/HTCondorAnnexUsersGuide.html

S U M M I T
Elasticity – Machine Learning & Natural Language Processing
at Clemson University, 1.1 Million vCPUs with EC2 Spot Instances
https://aws.amazon.com/blogs/aws/natural-language-processing-at-
clemson-university-1-1-million-vcpus-ec2-spot-instances/

S U M M I T
Children's Hospital of Philadelphia And Edico Genome Achieve Fastest-Ever
Analysis Of 1,000 Genomes
GUINNESS WORLD RECORDS title for Fastest time to analyze 1,000 human
genomes
https://www.prnewswire.com/news-releases/childrens-hospital-of-philadelphia-and-edico-
genome-achieve-fastest-ever-analysis-of-1000-genomes-300540026.html
The Amazon EC2 F1 instances, with
Xilinx Virtex UltraScale+ field
programmable gate arrays (FPGAs) was
used for 1,000 diverse pediatric genomes.
The study was completed in two hours
and twenty-five minutes.

S U M M I T
Accelerate Time to Science with NSF and Internet2 (E-
CAS)
AWS Collaboration with NSF: Several of them use Amazon F1 Instances (FPGAs)
https://www.businesswire.com/news/home/20190326005155/en/Internet2-National-Science-Foundation-Announce-Selection-First-Phase

S U M M I T
NSF’s Vision for a National
Cyberinfrastructure Ecosystem & Cloud
Services
Manish Parashar
Office Director
Office of Advanced Cyberinfrastructure,
Directorate for Computer & Information Science &
Engineering
National Science Foundation

S U M M I T
The National Science Foundation (NSF) is an independent federal agency created by Congress in 1950 "to promote the progress
of science; to advance the national health, prosperity, and welfare; to secure the national defense..." NSF is vital because we
support basic research and people to create knowledge that transforms the future.
Source: NSF at a Glance (https://www.nsf.gov/about/glance.jsp)
RESEARCH IDEAS
Windows on the
Universe:
Multi-messenger
Astrophysics
Quantum
Leap:
Leading the
Next
Quantum
Revolution
Navigating
the
New Arctic
Understanding
the Rules of
Life:
Predicting
Phenotype
PROCESS IDEAS
Mid-scale
Research
Infrastructure
Growing
Convergence
Research at
NSF
NSF 2026
NSF INCLUDES:
Enhancing STEM
through Diversity
and Inclusion
Harnessing
Data for 21st
Century
Science and
Engineering
Work at
the HT
Frontier

S U M M I T
NSF Office of Advanced Cyberinfrastructure (OAC)
Directorate for Computer & Information Science & Engineering (CISE)
$224M
FY 2018
research
budget
950
proposals
305
awards
32%
Success
Rate
People, organizations,
and communities
Data
Infrastructure
Gateways, Hubs,
and Services
Cloud
Resources
& Services
CI-Enabled
Instrumentation
Computing
Resources
R&E Networks,
Security Layers
Coordination
& User support
Software and
Workflow Systems
Pilots,
Testbeds
Source: https://dellweb.bfa.nsf.gov/starth.asp
Foster a cyberinfrastructure ecosystem to
transform science and engineering
research…
… through Research CI and CI research

S U M M I T
21st Century S&E Research Paradigms are Evolving
 Diverse / disruptive technologies
 Novel paradigms / Increasing role of clouds /
Growing capabilities & capacities at the edges
 Role of (non-traditional) software in taming
complexity
 Heightened emphasis on robust results
 Data-driven; Compute/data intensive
• Streaming data from observatories,
instruments
• Increasing use of ML
 End-to-end; collaborative
 Complex, dynamic workflows
Our cyberinfrastructre ecosystem must evolve….
How do we catalyze a Cyberinfrastructure Continuum from sensors to science, and reduce barriers to
CI adoption across research…?
TheoreticalExperimental Computational Data

S U M M I T
“…. an agile, integrated, robust, trustworthy and sustainable CI
ecosystem that drives new thinking and transformative
discoveries in all areas of S&E research and education”
6
Transforming Science Through
Cyberinfrastructure
NSF’s Blueprint for a National Cyberinfrastructure
Ecosystem for
Science and Engineering in the 21st Century
http://go.usa.gov/xm8bU

S U M M I T
A new vision…
Overarching principles:
View CI more holistically ~ CI continuum seamlessly integrating a spectrum of
resources, tools, services, and expertise to enable transformative discoveries.
Support translational research ~ core innovations  development of
community tools and frameworks  deployment and operation of sustainable
production CI.
Balance innovation with stability ~ longer continuity in production
computational capacity while fostering innovation and transition to production.
Couple discovery and CI innovation cycles ~ more rapidly address new
challenges and opportunities in an era of disruptive technologies and evolving
science needs.
Improve usability ~ ease pathways for discovering, accessing, understanding
and using powerful CI capabilities and services to enhance researcher
productivity and scientific impact.
An agile, integrated, robust, trustworthy and sustainable CI ecosystem that
drives new thinking and transformative discoveries in all areas of S&E
research and education.

S U M M I T
Computational blueprint.
Two strategies
Deploy a balanced computational ecosystem that supports
broad and diverse requirements, users and usage modes.
Leadership Class Systems, Capacity Systems, Federated
Resources, Prototypes and Testbeds, in concert with ongoing
investment in Campus CI and a new emphasis on inclusion of
emerging Cloud resources and services.
Achieve maximal impact from the array of computational capabilities
and expertise
Strategic investments in crosscutting coordination, resource
allocation, user services and support, performance measurement
capabilities, and CI workforce development.
Implement extensions and enhancements to
current investments and new programs and
opportunities in 2019 and beyond.
First of several blueprints focused on different elements of the CI ecosystem

S U M M I T
Computation for the Endless Frontier
29
Early user
access in
May 2019
Frontera will be:
• A leadership-class computational instrument with the broadest utility for all of S&E applications
• The largest CPU system on a US academic campus
• A national asset that complements other leadership-class computing investments in the US research ecosystem
https://www.tacc.utexas.edu/systems/frontera

S U M M I T
Clouds and the NSF CI Ecosystem

S U M M I T
NSF/CISE Cloud Access Program
 “Data- and compute-intensive research and education
efforts are benefiting from access to cloud computing
platforms, which provide robust, agile, reliable, and
scalable infrastructure”
 Efficiencies, services, scale not possible with per-
PI/campus clusters
 Cloud use already budgeted, particularly among
CISE researchers
 Current state: individual PIs or (in some cases)
institutions contract with cloud provider(s)
 no sharing of knowledge, resources (e.g., software,
data, ed.) among community
 Multiple reports NASEM[2016], OAC[2017],
Microsoft[2017], CISE[2018] cite importance of cloud and
suggests NSF cloud strategy needed
 Cloud Access: innovative pilot with CISE researchers

S U M M I T
Cloud Access
entity
Cloud Service Provider 1
Cloud Service Provider N
…
CISE-funded PI 1
(needing cloud resources)
CISE-funded PI M
…
“The Cloud Access entity will primarily serve PIs of participating CISE programs by
providing access to cloud resources and other services, and all CISE researchers and
educators with strategic technical guidance and training in using the cloud”

S U M M I T
Cloud Access
entity
“The cloud-facing functions include establishing relationships with public cloud
computing providers; establishing a structure for account management and resource
allocations; and engaging in strategic planning for use of public cloud computing
resources by the CISE community”
…
CISE-funded PI 1
CISE-funded PI M
…

S U M M I T
Cloud Access
entity
“The community-facing functions include providing user support specifically related
to the use of cloud computing resources; providing training and education support
related to cloud usage; and providing advice and strategic technical guidance about
the use of cloud computing resources in research and education projects.”
…
CISE-funded PI 1
CISE-funded PI M
…

S U M M I T
Cloud Access
entity
Entity: Works with cloud providers – “establishing partnerships with the various public
cloud computing providers.” Works with research community to design “user training
and other support to CISE researchers and educators using cloud computing in their
work” and “technical guidance for CISE researchers and educators interested in
using public cloud computing platforms.”
…
CISE-funded PI 1
CISE-funded PI M
…

S U M M I T
Cloud Access
entity
1: CISE PI requests cloud usage funds in standard proposal
2: As part of award process, CISE PD and Cloud Access entity interact to determine
cloud allocation credit
3: PI proposal funded, Supplement for PIs cloud service provided to entity, as needed
on top of initial $750K tranche.
1
…
CISE-funded PI 1
CISE-funded PI M
…
2 33

S U M M I T
Cloud Access
entityCloud Access
entity
4: PI, student training, interaction (may have happened earlier)
5: PI uses cloud resources
6: Cloud providers are paid via the entity (Maximized funds needed for research)
…
CISE-funded PI 1
CISE-funded PI M
…
5
6 4

S U M M I T
ECAS - Exploring Clouds to Accelerate Science
A competitive process
o Phase 1 = Using credits, Cloud Providers support 6 teams
(selected from proposals) to develop and demonstrate
computational science at scale.
o A panel of academic reviewers select best 2 in terms of
ACCELERATION of SCIENCE and INNOVATION
o Phase 2 = 2 awards of $500k + staff + F&A
Objectives
– Test effectiveness of commercial cloud for large
scale research
– Use accelerated hardware such as FPGA and
GPUs
– Explore Cloud AI and Machine Learning
frameworks
– Explore cloud provisioning and management of
resources.
– Examine performance metrics and identify gaps
Progress/Timeline
– Announced at SC18
– RFP issued Dec 2018
– Phase 1 - Mar 2019
– Phase 2 - Jul 2020
– Complete Sept 2021
Jamie Sunderland - jsunderland@internet2.edu
$100k
2x$500k
2x $380k
Acc Inov
$10
0k
6x
$100k
6
6x $80k 1 Year
1 Year
NSF Award #190444 under a Cooperative Agreement
with Internet2 : $3,030,955; 11/2018-10/2021
Phase 1
Phase 2

S U M M I T
The proposals selected for Phase 1 are:
“Accelerating Science by Integrating Commercial Cloud Resources in the
CIPRES Science Gateway”, Mark Miller, San Diego Supercomputing Center
(UCSD).
“Investigating Heterogeneous Computing at the Large Hadron Collider”,
Phillip Harris, Massachusetts Institute of Technology (MIT).
“Ice Cube computing in the cloud”, Benedikt Riedel, University of Wisconsin.
“Building Clouds: Worldwide building typology modelling from images”,
Daniel Aliaga, Purdue University.
“Deciphering the Brain's Neural Code Through Large-Scale Detailed
Simulation of Motor Cortex Circuits”, William Lytton, State University of New
York (SUNY Downstate MC)
“Development of BioCompute Objects for Integration into Galaxy in a Cloud
Computing Environment”, Raja Mazumder, George Washington University.
AWS + Nvidia V100 GPUs,
Bursting from XSEDE Comet
AWS FPGAs + Machine Learning
Framework
AWS FPGAs, GPUs + Tensor Flow
Machine Learning.
Computer Vision, Procedural
Modelling and ML
Uses NetPyNE and Slurm to burst
from Campus HPC up to 50,000
cores.
AWS and Direct Connect to
interconnect campus HPC and
Galaxy service on AWS.
The provider technologies

S U M M I T
Source: Weather Channel
Machine Learning for Improving Disaster Management and Response
Session ID: 301069 - Artificial Intelligence and Machine Learning in Research

S U M M I T
Research for Social Good (Collaboration: AWS, NSF and University of Nevada)
https://www.unr.edu/nevada-today/news/2019/big-data-wildfires

S U M M I T
Conclusion
• Science and society are being transformed by compute and data
– a connected, robust and secure cyberinfrastructure ecosystem is
essential
• Rapidly changing application requirements; resource and technology
landscapes
– Our cyberinfrastructure ecosystem must evolve in response
• Cloud services have to be an integral part of the cyberinfrastructure
ecosystem

S U M M I T
Join the conversation
• OAC Webinar Series
• 3rd Thursday @ 2PM ET
• OAC Newsletter
• OAC Townhalls (CASC, LFW, PEARC,
SC)
• Follow us on Twitter @NSF_CISE
Get involved
Reviews proposals, serve on panels
Visit NSF, get to know your programs
and Program Officers
Participate in NSF workshops and
visioning activities
Join NSF: serve as Program Officer,
Division Director, or Science Advisor
Stay informed
• Join the OAC, CISE Mailing Lists
• Learn about NSF events, programs,
webinars, etc.
• Send email to:
• oac-announce@listserv.nsf.gov
• cise-announce-subscribe-
request@listserv.nsf.gov

S U M M I T
Demián Arancibia
Astroinformatics Program Director
Chilean Ministry of Economy

S U M M I T
Observations Theory
Tycho Brahe
1546 - 1601
Johannes kepler & Isaac
Newton
1571-1630 & 1643-1727
Lives of Eminent and Illustrious Englishmen, 1830Jacques de Gheyn II, 1585 Public Domain

S U M M I T
Eagle Simulation, Institute for Com. Cosmology
2016
Sloan Digital Sky Survey
2012
Observations Theory

S U M M I T
credit: Ibsen (ALMA) Santander-Vela (SKAO)

S U M M I T
credit: Marca Chile

S U M M I T
C T A
V L T
G M T
L S S T
A L M A
S I M O N S
G E M I N IN T T
L C O
A C T
A P E X L C T
C C A T P
Milky Way In Multiple Wavelengths
S O A R E E L T
credit: The NRAO

S U M M I T
L S S T
And in real-time
credit: Matt Molloy

S U M M I T
Credit: Eagle Simulation, 2016
C T A
V L T
G M T
L S S T
A L M A
S I M O N SG E M I N IN T T
L C O
A C T
A P E X L C T
C C A T P
S O A R E E L T
How did our universe
form and grow?

S U M M I T
What is made of?
Dark Energy
71.4%
Dark Matter
24.0%
Known Matter
(Atoms)
4.6%
C T A
V L T
G M T
L S S T
A L M A
L C O
A C T
A P E X L C T
C C A T P
S O A R E E L T

S U M M I T
Equation: Drake, 1961
Image: University of
Rochester
Are we
alone?
C T A
V L T
G M T
L S S T
A L M A
L C O
A C T
A P E X L C T
C C A T P
S O A R E E L T

S U M M I T
A L M A
Credit: The NRAO

S U M M I T
C T A
V L T
G M T
A L M A
S I M O N S
G E M I N I
N T T
L C O
A C T
A P E X
L C T
C C A T P
S O A R
E E L T
L S S T

S U M M I T Credit: Astroinformatics Initiative

S U M M I T
Virtual Observatory Vision (Szalay, Djordovsky, Quinn & many others from 2000 on)
Credit: Astroinformatics Initiative

S U M M I T
Amazon Web Services based Astronomy (Data Observatory, ~2020)

S U M M I T
AWS-Based Astronomy (Data Observatory, ~2020)
Bring all computing, including data
access, visualization, and scientific
collaboration to the data

S U M M I T
Healthcare
Climate
Change
SCADA systems
Agricultur
e
Smart Cities
Commerce

S U M M I T
eBird collects data that are used to
estimate the distribution, abundance,
and trends of bird populations by
collaborating with a global network of
bird enthusiasts who submit their
observations to a central data archive.
Chris Wood
Cornell Lab of Ornithology
Cornell University

S U M M I T
Why birds?
• Found everywhere
• Easily detectible
• Indicators of environmental
health
• Engage millions of people

S U M M I T
Quick Overview
• More than . . .
• 645 million observations
• 46 million checklists
• 5.3 million locations
• Every country in the world
• 10,400 species
• 450,000 people have submitted data
• 250 peer-reviewed publications in last 5 years

S U M M I T
More than 45 million hours of effort in the field

S U M M I T
20% per year

S U M M I T

S U M M I T
1. Uneven sampling over space and time
2. Uneven detectability / identification
3. Uneven observation skill across participants
Observation process biases

S U M M I T
Data: Abundance,
Habitat, and Trends
~75 GB
Add Covariates
50k Core Hours
MODIS 100 GB
Query, Zero-fill
Modeling
~7k Core Hours
~1 TB Intermediate
eBird Reference Dataset
10k species
2004-2018
1 TB
Post-process
500 Core Hours
Convert, format, extract
100 GB
Status & Trends Analysis
Per species @ 2.8km x 2.8km x 1wk
eBird Reference Dataset
Registry of Open Data
122 species
Web Visualizations
122 species

S U M M I T
BirdReturns | Dynamic conservation in California

S U M M I T
Shorebird
Abundance
P( Surface Water ) Conservation Value
+ =
Reiter et al. 2015 Golet et al. 2017 in press
BirdReturns | Dynamic conservation in California

S U M M I T
Thank you!
S U M M I T
Sanjay Padhi
spadhi-aws@amazon.com

S U M M I T
S U M M I T

Accelerating Time to Science Using Cloud

Recommandé

Recommandé

Contenu connexe

Tendances

Tendances (6)

Similaire à Accelerating Time to Science Using Cloud

Similaire à Accelerating Time to Science Using Cloud (20)

Plus de Amazon Web Services

Plus de Amazon Web Services (20)

Accelerating Time to Science Using Cloud