application optimized performance: choosing the right instance (cpn212) | aws re:invent 2013
DESCRIPTION
(Presented by Intel) Each application places a different set of requirements on the underlying infrastructure. Whether it is web, big data analytics, technical computing, or general enterprise applications, applications are run more efficiently when performance, IO bandwidth, and memory capacity have been custom-tailored for that specific application. Jason Waxman, GM and VP of Intel’s Cloud Platform Group, looks under the hood at the different types of processors that comprise Amazon Web Services instances and shares insights from Intel IT and industry best practices for right-sizing infrastructure for different application characteristics and capabilities. By leveraging the underlying performance, security capabilities, and flexibility of various instance types, developers can more easily migrate applications into the cloud and drive down TCO for cloud-based services.TRANSCRIPT
Delivering Compelling Experiences: Choosing the Right Instance for Application Optimized Performance
Jason Waxman, GM & VP Cloud Platforms Group, Intel Corporation
November 14, 2013
Voice & Gestures Personal assistant
Natural Interaction
20X growth in speech driven mobile network traffic1
>22X in smartphones with gesture recognition features1
Perceptual
Compelling
Experiences drive Growth
Pervasive
Video/Media Content Delivery
Video Search
16X in mobile video traffic2
4X in servers for media/graphics3
Personal
Predictive Analytics Improve healthcare
43% CAGR Big Data & Analytics
Infrastructure4
1.http://www.digitalservicecloud.com/resources/blog/good-customer-service.html
On an average consumers tell:
about bad
experiences1
9 16 ..and
The High Costs of a Bad Experience
people about
good experiences…
Delivering Compelling Experiences
What’s
Required?
Diverse Needs, Common Themes CONSUMER
EXPECTATIONS
DEVELOPER
REQUIREMENTS
AGILITY
RELIABILITY
EFFICIENCY
Personal &
Customized
Service Availability
Privacy/Security
Cost Effective
Fast Service Delivery
Elasticity/Scalability
Stable, Consistent
Privacy/Security
Cost to Serve - Services & APIs reduce
headcount required
Delivering the best experience
SCALE: TAKE ADVANTAGE OF
AWS ELASTICITY
OPTIMIZE: CHOOSE THE RIGHT
INSTANCES
SECURE: MANAGE RISKS FOR
INCREASED DURABILITY
AGILITY EFFICIENCY RELIABILITY
Delivering the best experience
SCALE: TAKE ADVANTAGE OF
AWS ELASTICITY
OPTIMIZE: CHOOSE THE RIGHT
INSTANCES
SECURE: MANAGE RISKS FOR
INCREASED DURABILITY
AGILITY EFFICIENCY RELIABILITY
An Intel Company
Intel®
Cloud
Services
Platform
Intel is using Amazon Web Services
Mashery API Management Service
• 175 Customers
• 60,000 Applications
• 215,000 Developers
• 500,000,000 API calls/day
An Intel Company
Scaling enables responsiveness Mashery relies on AWS elasticity
• Capacity Planning - Robust Development and QA
environment to perform load tests and deploy proof-
of-concept silos
• Modular Infrastructure - Loosely coupled single-
purpose servers that scale horizontally
• Globally Distributed - Extend the infrastructure to
where you customers are…every AWS Region,
reaching every corner of the globe.
From 100 queries/sec to 100,000 queries/sec in a matter of minutes
NEWS VOLUMES FOR SOME
CUSTOMERS AT PASSING OF
CELEBRITY IN EARLY 2012
100X
Kevin Baillee CEO, Atomic Fiction
Company Overview • Atomic Fiction crafts high-end visual effects (VFX) for film and television.
• Specialties include digital environments and character work
• Staff scales with projects, varying between 15 and 50 artists
• Known for high end work, medium volume, low cost
• Big company infrastructure, small company vibe
• Developing innovative approaches to reducing technological costs in
order to fee up resources for experienced artistic talent.
Company Overview
Our AWS Story • Pixar-sized render farm in minutes. iMac-sized the next.
• Only pay for what we use
• No physical limits on computing = no physical limits on creativity
• Seamless experience through tools like ZYNC
• Decreased load, thus increased performance, on local filesystem
• Same price for faster artist turnaround
• 100 computers for 10 hours = $2,200
• 1000 computers for 1 hour = $2,200
Problems we’re trying to solve • The problem: how do we “unlimit” creativity while staying profitable?
• Freeing the creative so that directors can achieve their vision
• Artists need iterations in order to hit “the look” and stay on schedule
• Need security, task-appropriate stats, and unlimited availability on demand
• Choosing the right instance: speed vs memory vs cost
• c1.xlarge – 20 compute units, 7GB RAM
• Low cost per hour for lightweight compositing tasks
• cc2.8xlarge – 88 compute units, 60.5GB RAM
• Beefy RAM, good $ per compute unit cost proposition
• Working with our partners at ZYNC, implementation was plug & play!
Star Trek Into Darkness
Star Trek Into Darkness
Star Trek Into Darkness
Star Trek Into Darkness
Key Learnings/What’s Next? • For Star Trek Into Darkness, we achieved exactly what JJ Abrams wanted
• Key findings:
• For over 80% of tasks, cc2.8xlarge was fastest & most cost effective
• Given higher per-hour costs, efficient utilization of high end instances is
critical. Partial hours = wasted money!
• Burstability was critical for hitting deadlines. Ran between 0 and 400
instances simultaneously depending on the needs of the moment.
• Grew over 200% month-over-month two months in a row
• Our ideal instance would be inexpensive, high compute power (Intel Xeon
E5-2600 v2), medium memory (32-48GB) with nVidia GPU.
• Next for us: moving even more of our workflow into the cloud!
Delivering the best experience
SCALE: TAKE ADVANTAGE OF
AWS ELASTICITY
OPTIMIZE: CHOOSE THE RIGHT
INSTANCES
SECURE: MANAGE RISKS FOR
INCREASED DURABILITY
AGILITY EFFICIENCY RELIABILITY
Optimize: Choose the right Instance
E- Commerce
Dedicated
Hosting
Enterprise
Applications
High
Performance
Computing
Big Data Content Delivery
and Gaming
Graphics
Rendering
I/O Intensive
CP
U &
Mem
ory
In
ten
siv
e
Cold
Storage
Low End
Networking
Edge
Routing
Storage
De-dupe
Cloud RAN
Small
Cell
Higher latency, lower throughput Lower latency, higher throughput
Micro Instance
M3 Standard
Instance
E5-2670
Cluster
Compute
Instance
E5-2670 Cluster Graphics Instance
X5570
M1 Standard Instance
C1 Compute
Instance M2 Memory Optimized
CR1 Memory
Optimized
E5-2670
Storage –
Optimized
E5-2650
High Memory
G2 GPU Instance
E5-2670
Intel® Cloud Services Platform
• 175,000 Users
• 5M users by 2014
• Entire path of production
is on AWS
• 1448 instances…
Our Wake Up Moment… One month we spent $300K…60% of which we found later was wasted…
• We were spinning up instances and forgetting they were on
• We had larger instances than we actually needed
• Most instances never went over 10% utilization…
PER WEEK OF
UNUTILIZED INSTANCES
108hrs
OF SAVINGS BY TURNING
OFF NON PRODUCTION
INSTANCES AFTER HOURS
$100K
TOTAL SAVINGS
>60%
Optimize for Efficiency Select the Right Instance
Keys to Success • Analyze: “Trusted Advisor”
• Select the right type of instance
for your workload
• Size & Features
• # of Instances
• Reserve instances where possible
for cost efficiency
Steve Litster, PhD. Global Head of Scientific Computing
Novartis Institutes for Biomedical Research
Accelerating Science
Novartis Institutes for BioMedical Research
(NIBR)
Unique research strategy driven by patient needs
World-class research organization with about
6000 scientists globally
Intensifying focus on molecular pathways shared by
various diseases
Integration of clinical insights with mechanistic
understanding of disease
Research-to-Development transition redefined
through fast and rigorous “proof-of-concept” trials
Strategic alliances with academia and biotech
strengthen preclinical pipeline
Requirements
Large Scale Computational Chemistry Simulation
Results in under a week
Flexible target
Ability to run multiple experiments “on-demand”
Challenges
Sustained access to 50000+ compute cores
Ability to monitor and re-launch jobs
No additional Capital Expenditure
Internal HPCC already running at capacity
Job Profile
Embarrassingly Parallel
CPU Bound
Low I/O, Memory and Network requirements
Accelerating the Science
Virtual Screening
Target
Molecule Compound
Molecule
binding
site
"Lock" "Keys"
The Cloud: Flexible Science on Flexible Infrastructure
Engineering the right infrastructure for a workload:
Software runs the same job many times across instance types
Measures the throughput and determines the $ per job
Use the instances that provide the best scientific ROI
CC2 instance (Intel Xeon® ‘Sandy Bridge’) ran best for this
Metric Count
Compute Hours of Science 341,700 hours
Compute Days of Science 14,238 days
Compute Years of Science 39 years
AWS Instance Count-CC2 10,600 instances
Super Computing in the Cloud
$44 Million infrastructure
10 million compounds screened
39 Drug Design years in 11 hours for a cost of …$4,232
3 promising compounds identified
Key Learnings/What’s Next?
Diversity of Life Sciences brings unique challenges
Spend the time analyzing and tuning
Flexibility, Scalability and Performance
Time to rethink and retool
Challenge the Science and the Scientist
Collaboration
Future plans
Chemical Universe : 166 Billion cpds ≤ 17 atoms (Extreme scale CPU)
Next Generation Sequencing in the Cloud (Extreme CPU, Mem, I/O)
“Disruptive” Technologies-Imaging (x10 NGS requirements!)
Delivering the best experience
SCALE: TAKE ADVANTAGE OF
AWS ELASTICITY
OPTIMIZE: CHOOSE THE RIGHT
INSTANCES
SECURE: MANAGE RISKS FOR
INCREASED DURABILITY
AGILITY EFFICIENCY RELIABILITY
Optimizing for Security Performance higher performance saves cost
Intel Internal Benchmark
Instance Requirements for 400mbps
OPEN SSL PERFORMANCE
1 2 3
m1.x
larg
e
42 $10K/year
m3.x
larg
e
(w/A
ES
-NI)
1 $5.6K/year 21
1 - Not required but added for redundancy
2 - Requirement is 3.2, but you can’t buy .2, so round up to 4
SAVINGS
50-75% By upgrading from
m1.xlarge to the more
expensive m3.xlarge
because of AES-NI
NASDAQ OMX
36
WE LIST ~3300 GLOBAL COMPANIES WORTH
IN MARKET CAP REPRESENTING
$6 TRILLION
DIVERSE INDUSTRIES AND
MANY OF THE WORLD’S
MOST WELL-KNOWN AND
INNOVATIVE BRANDS
OUR TECHNOLOGY
IS USED TO POWER MORE THAN
I N 5 0
C O U N T R I E S
70 MARKETPLACES
OUR GLOBAL PLATFORM
CAN HANDLE MORE THAN
1 MILLION MESSAGES/SECOND AT A MEDIAN SPEED OF
S U B - 5 5
M I C R O S E C O N D S
FinQloud R3 (Regulatory Record Retention)
Security
Elastic Durable/Available
Cost Effective Transparent
POWERED BY AMAZON WEB SERVICES
Cloud Computing Platform
Exclusively for Financial Services
NASDAQ OMX Security Protocols
* Highly confidential data must be encrypted
and the keys must be stored in HSMs
Data Classification Encryption
Audit
AWS Built In Security
Security
at all times (in flight and at
rest) SSL for all data
Define and enforce what is
and is not approved for
the Cloud
Any action someone does
in R3 is audited and
fully transparent to system
admins and regulators
IAM, MFA, VPC, Direct
Connect private circuits,
routing/firewalls, etc
Data Classification
Technology: What’s Next
Intel® Xeon® Processor
E5-2600 v2 Family
Software Defined Infrastructure
Rack Scale Architecture
Diversity of Datacenter Workloads
E- Commerce
Dedicated
Hosting
Enterprise
Applications
High
Performance
Computing
Big Data Content Delivery
and Gaming
Graphics
Rendering
I/O Intensive
CP
U &
Mem
ory
In
ten
siv
e
Cold
Storage
Low End
Networking
Edge
Routing
Storage
De-dupe
Cloud RAN
Small
Cell
Higher latency, lower throughput Lower latency, higher throughput
Micro Instance
M3 Standard
Instance
E5-2670
Cluster
Compute
Instance
E5-2670 Cluster Graphics Instance
X5570
M1 Standard Instance
C1 Compute
Instance M2 Memory Optimized
CR1 Memory
Optimized
E5-2670
Storage –
Optimized
E5-2650
High Memory
G2 GPU Instance
E5-2670
“New” EC2 C3
Compute Optimized
w/ Latest Intel® Xeon®
E5-2600v2 Processors
E5-2680v2
“New” EC2
Storage Optimized
w/ Latest Intel® Xeon®
E5-2600v2 Processors
E5-2670v2
Technology Matters
Amazon Web Services
discloses instances based on
Intel Xeon
New 2013 AWS C3 Compute-Optimized
Instance
Powered by “NEW” Intel® Xeon® E5-2600v2 processor family
26K cores based on Intel® Xeon® processor E5-2680v2
SC’13
484 TFLOPs*
*SC’13 Submission
Platform Flexibility - Increase
useful life, and capacity
The Future of the platform: Intel Rack Scale Architecture Innovation
Orchestration
Increases Agility, Efficiency & Reliability
CPU / Mem Modules
Silicon – Intel® Atom™ & Xeon
Photonics & switch fabric
Storage – PCIE –SSD &
Caching
Open Network Platform Network platform – Flexible &
Cost effective
Increase utilization thru storage
aggregation
Extreme Compute and Network
bandwidth
The Future of Infrastructure: Intel’s Approach to Software Defined Infrastucture
EXPOSED & INTEGRATED TELEMETRY
Hardware and infrastructure attributes are exposed
and integrated with orchestration software for deeper
insight & optimal provisioning management
BROADEST ENABLED ECOSYSTEM
Integrated and optimized for all leading commercial
and open source operating environments for more
seamless Data Center operations
Rack Scale Architecture
• Scalable Intel®
Atom® &
Xeon® storage
solutions
• SSD’s with
Cache
Acceleration
• Luster
• NVM/Crystal
Ridge
• Open Network
Platforms
• Wind River
OS
• DPDK
• Cave Creek
• Silicon
Photonics
• Intel Xeon
• Atom®C2000
• Intel Xeon
Phi
• Integrated
graphics
• TXT
AMAZON WEB SERVICES (AWS)
Storage Network Compute
Service Assurance Manager
PLATFORM AND ARCHITECTURAL LEADERSHIP
Standards-based compute, network and storage
building blocks in Intel’s Rack Scale Architecture drive
maximum infrastructure efficiency and flexibility
A world where the application defines the system
Delivering the best experience
SCALE: TAKE ADVANTAGE OF
AWS ELASTICITY
OPTIMIZE: CHOOSE THE RIGHT
INSTANCES
SECURE: MANAGE RISKS FOR
INCREASED DURABILITY
AGILITY EFFICIENCY RELIABILITY