High Performance Computing on AWS

© 2018, Amazon Web Services, Inc. or its Affiliates. All rights reserved
Pop-up Loft
High-Performance Computing on AWS
Francesco Ruffino,
Sr. Global Specialized HPC Solution Architect
fruffino@amazon.com

Motivations – why HPC in the Cloud?

On-premises operational challenges
• Difficult to match variable demand to static on premise compute
– Over-provision vs under-provision
• Right-sizing
– Hard to experiment with different HW, leverage latest models, refresh cycles
• Environment Management
– Managing BCP/DR
– Infrastructure operations (separate for compute, network, storage, OS,..)
– Chargebacks – hard in multi-tenant model to track resource usage/metering;
lower usage apps get penalized

High Performance Computing on AWS
• Innovate faster with virtually unlimited infrastructure
enabling scaling and agility not attainable on-premises
• Optimize cost with flexible resource selection and pay
per use
• Increase collaboration with secure access to
clusters around the world
Faster Time to
Results
Better ROI

AWS Advantages for Multiple HPC Workload Types
Tightly Coupled
Parallel Computing
Loosely Coupled
Parallel Computing
Accelerated
Computing
Visualization and
Interpretation
High Performance
Data Storage and
Analytics
Scale
EC2Spot
Pricing
Early Access to
Technology
Choice Performance
Deriveunique
insights withAI/ML
SkiptheQueue Viewresults
instantly

Amazon EC2 Instances
General
purpose
Dense
storage
Compute
optimized
FPGA
GPU
Compute
Storage
optimized
Graphics
intensive
Memory
optimized
P2M4 D2 X1 G2T2 R4I3C5 F1M5 P3H1 EC2 Bare MetalG3T2 Unlimited X1eI2C4
High
I/O
General
purpose
burstable
Direct access
to physical
server
resources
ü Select compute that best fits the workload profile
ü Match the architecture to the job, not viceversa
ü Optimize price/performance of your HPC Workloads with widest range
of compute instances
ü Benefit from the AWS pace of innovation

Cost advantages
On Premises
Capital Expense Model
Amazon Web Services
Pay As You Go Model
• Use only what you need
• Multiple pricing models
• High upfront capital cost
• High cost of ongoing
support
8

Networks
• AWS proprietary networking
– Full bi-section bandwidth in placement groups
• Elastic Network Adapter (ENA)
– Supports network speeds of up to 25 Gbps in placement groups
– Multi-queue support
• VPC (Network segregation)
• Direct Connect (1/10GigE)

Important enablers for HPC on the cloud
• Compute performance – CPUs, GPUs, FPGAs
• Memory performance – high RAM requirements in many applications
• Network performance – throughput, latency, and consistency
• Storage performance – including shared filesystems
• Automation and cluster/job management
• Remote graphics for interactive applications
…and SCALE

Defining HPC – example use cases
Data Light
Minimal
requirements
for high
performance
storage
Data Heavy
Benefits from
access to high
performance
storage
Clustered (Tightly coupled)
Distributed / Grid (Loosely coupled)
• Fluid dynamics
• Weather forecasting
• Materials simulations
• Crash simulations
• Risk simulations
• Molecular modeling
• Contextual search
• Logistics simulations
• Animation and VFX
• Semiconductor verification
• Image processing/GIS
• Genomics
• Seismic processing
• Metagenomics
• Astrophysics
• Deep learning

Cluster and grid HPC in the cloud
Cluster HPC
Tightly coupled, latency-
sensitive applications
Use larger EC2 compute
instances, placement
groups, enhanced
networking, HPC job
schedulers
Grid HPC
Loosely coupled, pleasingly
parallel
Use a variety of EC2
instances, multiple AZs,
Spot, Auto Scaling, Amazon
SQS, AWS Batch
Grids of Clusters
Running parallel cluster
jobs, parameter studies
Use a grid strategy on the
cloud to run a group of
parallel, individually-
clustered HPC jobs

Grid Computing Examples

1.1M vCPUs for machine learning
A group of researchers
from Clemson University
achieved a remarkable
milestone while studying
topic modeling, an
important component of
machine learning
associated with natural
language processing,
breaking the record for
creating the largest
high-performance
cluster in the cloud by
using more than
1,100,000 vCPUs on
Amazon EC2 Spot
Instances running in a
single AWS region.
Thegraphhighlights
theelastic, automatic
expansionof
resources.
Clemsontook
advantageof thenew
per-secondbillingfor
EC2instances.
ThevCPUcount
usageis comparable
tothecorecount on
thelargest
supercomputers inthe
world.
S3
P ro v is io n in
g a n d
w o rk flo w
a u to m a tio n
s o ftw a re
S3
J O B
S C R IP T
C LO U D Y
C LU S T E R
A P Is
L O G IN S C H E D U L E R
S LU R M
A U T O
S C A L IN G
S P O T F L E E T
C C Q
S3
DDB VPC
https://aws.amazon.com/blogs/aws/natural-language-processing-
at-clemson-university-1-1-million-vcpus-ec2-spot-instances/

HPC in design and manufacturing
Applications for engineering:
• Molecular dynamics, CAD, CAE, EDA
• Collaboration tools for engineering
• Big data for manufacturing yield analysis
Running drive-head simulations at scale
Millions of parallel parameter sweeps,
running months of simulations in just hours
Over 85,000 Intel cores running at peak,
using Spot Instances

Cluster Computing Examples

Fluid dynamics – Ansys Fluent
• C4.8xlarge instance type
• 140M cell model
• F1 car CFD benchmark
https://www.ansys-blog.com/simulation-on-the-cloud/

HPC in aerospace
Boom leverages Rescale andAWS to enable supersonic travel
“Rescale’s ScaleX
cloud platform is a
game-changer for
engineering. It gives
Boom computing
resources comparable
to building a large on-
premise HPC center.
Rescale lets us move
fast with minimal
capital spending and
resources overhead.”
Josh Krall
CTO & Co-Founder
• Simulated vortex lift with 200M cell models on 512+ cores
• Increased simulation throughput: 100 jobs in parallel with 6x
speedup per job → 600x speedup
• Eliminated IT overhead, including server capital costs & in-house
IT and software teams
• Elastic HPC capacity and pay-as-you-go AWS clusters allow
business agility & ability to scale

Performance Considerations

Performance considerations
For tightly-coupled cluster workloads
Network
• Use a placement group
• Enable enhanced networking
MPI libraries
• Test with Intel MPI and OpenMPI 3.0,
and make use of available tunings
Domain decomposition
• Choose number of cells per core for
either pre-core efficiency or for faster
results
Test using real-world examples
• Use large cases for testing: do not
benchmark scalability using only small
examples
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
0.0
500.0
1000.0
1500.0
2000.0
2500.0
0 500 1000 1500 2000 2500 3000 3500 4000
Time (S)
Scale-Up
Cores
WRF 2.5 km CONUS Benchmark
Scale-Up time

Performance considerations
For all HPC workloads
OS version
• UseAmazon Linux or an updated 3.10+
kernel – 4.0+ if using NVME on F1 or I3
Instance types
• C5, C4, M5, R4 are the best choices
today – but always test with the latest
EC2 instances
Processor states
• Use P-states to reduce processor
variability
Hyper-threading and affinity
• Test with Hyper-threading (HT) on and
off – usually off is best, but not always
• Use CPU affinity to pin threads to CPU
cores when HT is off

AWS Storage is a platform
File Block Object
Amazon EFS Amazon EBS
Amazon EC2
Instance Store Amazon S3 / S3-IA Amazon Glacier
Data Transfer
INTERNET / VPN AWS DIRECT
CONNECT
AMAZON
CLOUDFRONT
S3 TRANSFER
ACCELERATION
ISV
CONNECTORS
STORAGE
GATEWAY
AWS
SNOWBALL
AMAZON KINESIS
FIREHOSE

EFS EBS + EC2 Amazon S3 Amazon Glacier
Highly available, multi-AZ,
fully managed network-
attached elastic file system.
For near-line, highly-
available storage of files in a
traditional NFS format
(NFSv4).
Create a single-AZ shared
file system using EC2 and
EBS, with third-party or
open source software (ZFS,
Weka.io, Intel Lustre, etc).
For near-line storage of files
optimized for high IOPS.
Secure, durable, highly-
scalable object storage.
Fast access, low cost.
For long-term durable
storage of data, in a readily
accessible get/put access
format.
Secure, durable, long term,
highly cost-effective object
storage.
For long-term storage and
archival of data that is
infrequently accessed.
Use for read-often,
temporary working storage
Use for high-IOPS,
temporary working storage
Primary durable and
scalable storage for critical
data
Use for long-term, lower-
cost archival of critical data
Optimize HPC storage

HPC Deployment and Automation
© 2018, Amazon Web3

33
Deploy multiple HPC clusters
Running at the same time, and tuned for each workload

HPC automation with CfnCluster
• CfnCluster simplifies
deployment of HPC in the
cloud, including integrating
with popular HPC schedulers
• Built on AWS CloudFormation,
easy to modify to meet specific
application or project
requirements

AWS Batch for HPC workloads
Fully Managed
No software to install or
servers to manage
AWS Batch provisions,
manages and scales your
infrastructure
Integrated with AWS
Natively integrated with
the AWS platform
AWS Batch jobs can
easily and security interact
with services such as
Amazon S3, DynamoDB,
and Rekognition
Cost-optimized
Resource
Provisioning
AWS Batch automatically
provisions compute
resources tailored to the
needs of your jobs using
EC2 and EC2 Spot

Graphics and collaboration with DCV and AppStream
36
Cloud can be used for pre-and post
processing as well as HPC
• Use GPUs in the cloud for remote
rendering and remote desktops
Cloud is more secure for collaboration
• Encrypt the data in flight and at rest
• Manage your own keys and
credentials
• Deliver pixels to your collaborators,
not the actual data

Deploying HPC on AWS
3DGRAPHICSVIRTUALWORKSTATION
LICENSEMANAGERSANDCLUSTERHEADNODESWITHJOBSCHEDULERS
CLOUD-BASED, AUTO-SCALINGHPCCLUSTERS
SHAREDFILESTORAGE STORAGECACHE
OnAWS, secure and
well-optimized HPC
clusters can be
automatically created,
operated, and torn down
in just minutes
AmazonS3
andAmazonGlacier
ON-PREMISES
HPCRESOURCES
CorporateDatacenter
AWSSNOWBALL
AWSDIRECTCONNECT
THIN- NOLOCALDATA-
ORZEROCLIENT
A P P S T R E A M 2 .0
A W S B A T C H

Pop-up Loft
aws.amazon.com/hpc
Thank you!

Pop-up Loft
High-Performance Computing on AWS
Financial Services applications on AWS
Francesco Ruffino,
Sr. Global Specialized HPC Solution Architect

Adoption in the Financial Services industry is
accelerating

New forms of risk are
constantly emerging.
Consumers have higher
expectations and more choices.
Data is revealing go-to-market
and cost-saving opportunities.
The Financial Services industry continues to evolve

Multiple forces are converging to drive cloud adoption in the
industry
Constant pressureonmargins
andresourcescarcity
Risingdatavolumesandneedfor cost-
effectiveanalyticstoenableAI/ML
Regulatoryoversight andever-expanding
reportingobligations
Recognitionof superior security/
dataprivacyinthecloud
Ongoingcompetition
fromnewentrants
Legacyprocesses/infrastructure
hinderinginnovation

Security and compliance are moving from
obligation to advantage

Compliance starts with AWS’ Shared Responsibility Model
Customer
Customer data
Operating system, network & firewall configuration
AWS Platform, applications, identity & access managementResponsibility for
security in the
cloud
Compute Storage Database Networking
Client-side data encryption
& data Integrity
authentication
Server-side encryption
(file system and / or
data)
Networking traffic
protection (encryption /
integrity / identity)
Responsibility for
security of the
cloud
AWSglobal
Infrastructure
Edge
Locations
Regions
AvailabilityZones

Customers rely on AWS’ compliance with global standards
Certifications &Attestations Laws, Regulations andPrivacy Alignments &Frameworks
Cloud Computing Compliance Controls Catalogue (C5) ! CISPE " CIS (Center for Internet Security) #
Cyber Essentials Plus $ EU Model Clauses " CJIS (US FBI) %
DoD SRG % FERPA % CSA (Cloud Security Alliance) #
FedRAMP % GLBA % Esquema Nacional de Seguridad &
FIPS % HIPAA % EU-US Privacy Shield "
IRAP ' HITECH # FISC (
ISO 9001 # IRS 1075 % FISMA %
ISO 27001 # ITAR % G-Cloud $
ISO 27017 # My Number Act ( GxP (US FDA CFR 21 Part 11) %
ISO 27018 # Data Protection Act – 1988 $ ICREA #
MLPS Level 3 ) VPAT / Section 508 % IT Grundschutz !
MTCS * Data Protection Directive " MITA 3.0 (US Medicaid) %
PCI DSS Level 1 + Privacy Act [Australia] ' MPAA %
SEC Rule 17-a-4(f) % Privacy Act [New Zealand] , NIST %
SOC 1, SOC 2, SOC 3 # PDPA - 2010 [Malaysia] - Uptime Institute Tiers #
PDPA - 2012 [Singapore] * Cloud Security Principles $
PIPEDA [Canada] .
# = industry or global standard Agencia Española de Protección de Datos &

Documentation
of controls relevant to
specific AW S services
Information
on AW S policies,
processes, and controls
This support includes easy access to our compliance documentation
Validation
that AW S controls are
operating effectively
What is it?
A globally available, no-cost portal that
provides on-demand access to AWS’
most recent external security and
compliance certifications
Global Certifications and Attestations
How does it work?
Customers can review reports, align AWS
controls to their own control frameworks,
and use the reports to verify that AWS
controls are operating effectively
The AWS Artifact tool supports increased transparency

AWS is the first choice for highly regulated organizations
Security
enhancements from
1M+ customer
experiences
AWS industry-
leading
security teams: 24/7,
365 days a year
Security infrastructure
built to satisfy military,
global banks, and other
high-sensitivity
organizations
Over 50 global
compliance
certifications and
accreditations
“We can be far more secure in the cloud and
achieve a higher level of assurance at a much
lower cost, in terms of effort and dollars
invested. We determined that securityinAWSis
superior toour on-premises datacenter across
several dimensions, including patching,
encryption, auditing and logging, entitlements,
and compliance.” – John Brady, CISO, FINRA

Grid Computing in Financial Services Today

Compute needs in financial services are changing
• Financial simulations are essential to the operations of all
Financial Services institutions (FSI) to identify and manage risk,
optimize capital, and make informed investment and pricing
decisions
• Regulatory bodies are requiring FSIs to perform higher levels of
stress testing to maintain adequate capital ratios, while regulatory
changes have increased the complexity of allocating capital and
collateral to meet margin and solvency requirements
• The development of new products and trading strategies,
particularly for complex products are leveraging a greater variety
of data sets increasing the complexity of design and back testing

Financial modeling has grown more onerous
Compute-intensive
calculations
• More granular risk factors
• Wider range of scenarios
• More historical data
Broad regulatory requirements
• Comprehensive Capital Analysis and
Review (Banking/Dodd Frank)
• Solvency Capital Requirements
(Insurance/Basel II)
• Fundamental Review of the Trading Book
(Insurance/Basel III)
Diverse risk analysis models
• Market risk
• Credit risk
• Liquidity risk
Large amounts of compute
resources needed to run simulations

The Challenge of Managing Grid Utilization

Schedule
impact!
High utilization can create backlogs
The Cluster as Seen by the Application User

Conflicting goals
• Grid users seek fastest possible time-to-results
• Grid workloads are variable not steady-state
• IT support team seeks the most economic, highest
possible utilization
Result:
• The job queue becomes the capacity buffer
• Job completion times are hard to predict
• Users are frustrated and run fewer jobs
• Innovation is throttled
?
And lengthy job queues can carry high costs
Large job

Solving the Problem Through Cloud-enabled Grids
Grid Reference Architectures

Reference Architectures – Grid Brokers
Typical for commercial Grid
Middleware:
• Symphony
• DataSynapse
• MSFT HPC Server
Compute Private Zone (Subnet/VPC)
Grid
Broker
Grid
Broker
Brokers
(Subnet/VPC)
Auto-Scaling Group
AWS REGION
Application Data (input/output), Configuration/Reference Data, Binaries
https://aws.amazon.com/
blogs/aws/creating-a-1-3-
million-vcpu-grid-on-aws-
using-ec2-spot-instances-
and-tibco-gridserver/
Grid
Client
Customer DataCenter

Grid reference architecture
Typical Lift-n-shift
deployment
• Home-grown
middleware or
commercial/open-
source schedulers
(Slurm, Sun
GridEngine, Univa…)
• Leverage shared
filesystem (GPFS,
Lustre, DIY NFS)
Virtual Private Cloud
Subnet Placement Group
10.40.0.0/16
10.40.10.0/20
Am azon S3
EFS
IAM Role
MSSNode
Scheduler
Node
Compute
Nodes
Compute
Nodes
Metadata
Servers
Datanode
Servers
Am azon
CloudW atch
AW S
CloudForm ation
AW S
CloudTrail
AW S
Config
AWS KMS

Reference Architectures: AWS Batch Managed Grid

AWS Batch: major features
• Array job: a single call runs as collection of related jobs across multiple hosts as
embarrassingly parallel jobs – ideal for Monte Carlo simulations, parametric
sweeps
• Job dependencies & Retries
• CloudWatch Events can trigger AWS Batch Jobs
• Customer provided AMIs option
• Support for per-second billing
• AWS Batch will evaluate compute resources more frequently and immediately
scale down any idle instances when no more runnable jobs in your job queues

Serverless: the Fannie Mae use case
[re:Invent 2017] Performance Computing for Financial Services with Serverless Compute:
https://www.youtube.com/watch?v=wLEHOTXU3As -- Implementing Monte Carlo simulations using map/reduce
approach

Lambda-based HPC platform
• Cost effective: never pay for idle. Cost based on actual vCPU us and not
processing capacity estate
• Performance improvement at zero cost: 1 Lambda x 15,000 hours = 15,000
Lambdas x 1 hour
• Shorter time to market: Ability to burst to cloud immediately to access
additional computing resources
• Maximize S3 performance by distributing key names to evenly distribute objects
across the partitions
• Set up separate AWS account for unlimited Lambda access without running out
of IP addresses

Pop-up Loft
aws.amazon.com/financial-services/grid-computing
Thank you

High Performance Computing on AWS

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Similar to High Performance Computing on AWS

Similar to High Performance Computing on AWS (20)

More from Amazon Web Services

More from Amazon Web Services (20)

High Performance Computing on AWS