S U M M I T - pages.awscloud.com · Accelerate application for fraction of cost of standalone...
Transcript of S U M M I T - pages.awscloud.com · Accelerate application for fraction of cost of standalone...
S U M M I TTo k y o
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
EC2 Innovation at Scale
Raj PaiDirector of Product ManagementAmazon EC2
H 1 - 0 3
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon EC2
Resources
Instances
Storage
Networking
Availability
Regions and AZs
Outposts
Management
Deployment
Monitoring
Administration
Purchase Options
On Demand
Reserved
Spot
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon EC2
Resources
Instances
Storage
Networking
Availability
Regions and AZs
Placement Groups
Load Balancing
Auto Scaling
Management
Deployment
Monitoring
Administration
Purchase Options
On Demand
Reserved
Spot
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon Elastic Compute Cloud (EC2) Virtual servers in the cloud
Physical servers in AWS global regions
Guest 1 Guest 2
Guest 1 Guest 2 Guest n
Hypervisor
Host server
EC2 instances
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon EC2 12+ years ago…
Scale up or down quickly, as needed
Pay for what you use “One size fits all”
M1
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Continued rapid pace of innovation Instance growth
instances
2007 2019
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Nitro Card Nitro Security Chip Nitro Hypervisor
Local NVMe storage
Elastic Block Storage
Networking, monitoring, and security
Integrated into motherboard
Protects hardware resources
Lightweight hypervisor
Memory and CPU allocation
Bare Metal-like performance
Innovation enabled by AWS Nitro System
Modular building blocks for rapid design and delivery of Amazon EC2 instances
Run virtualized instances with perf indistinguishable from Bare Metal
Run bare metal workloads on Amazon EC2 with all the elasticity, security, scale, and services of AWS
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon EC2 instance characteristics
M5d.xlargeInstance family
Instance generation
Instance size
Instance type
CPU
Memory
Storage
Network performance
*Additional capabilities
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon maintained
Broad set of Linux and Windows images
Kept up-to-date by Amazon in each region
Marketplace
maintained
Managed and maintained by AWS Marketplace
partners
Your machine images
AMIs you have created from Amazon EC2 instances
Can keep private, share withother accounts, or publish to
the community
Amazon Machine Images (AMIs)
Amazon Linux 2 with five years of long term support
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Choice of processors and architectures
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
General purpose instance workloads
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
M5: General purpose instances• Balance of compute, memory, and networking resources
• Powered by 2.5 GHz Intel Xeon Scalable Processors (Skylake)
• Largest instance size, m5.24xlarge has 96 vCPUs and 384 GiB of memory
• Improved network and EBS performance on smaller sizes
• Support for Intel AVX-512 offering up to twice the performance for vector and floating point workloads
14% price/performance improvement with M5
M4 M5
M5dSSD
M5a: Now available with AMD EPYC 7000 processor for 10% lower cost
M5d: Now available with high performance local NVMe SSD storage
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Low utilization
High utilization
Opportunity: Most instances aren’t very busy
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
T3: Burstable general-purpose instances
Balance of compute, memory, and network
Baseline level of CPU performance with the ability to burst CPU usage when needed at any time for as long as required
Lowest cost instance at $0.0052 per hour and up to 30% better price performance over T2
With T3 Unlimited bursting over baseline is only $0.05 per vCPU-hour, averaged over 24 hours
…7 sizes
T3a: Available with AMD EPYC 7000 processor for 10% lower cost
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
A1: First Arm instance in Amazon EC2
Optimized cost and performance for scale-out applications
Up to 45% cost savings AWS Graviton Processor with Arm-based cores and custom silicon
Lower cost for scale-out workloads
Arm-based development platform
Broad software and tool ing support
…6 sizes…5 sizes
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Choosing between Amazon EC2 General Purpose Instances
M5/M5a Instances
Balance of compute, memory, and network resources
T3 Instances
A1 Instances
Optimized for most workloads with occasionalhigh CPU use
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
R5: Memory optimized instancesMemory-optimized instances with 8:1 GiB to vCPU
2.5 GHz Intel Xeon Scalable Processors (Skylake)
Up to 25 Gbps NW bandwidth
R5d instances include up to 3.6 TB of local NVMe SSD
…6 sizes…6 sizes
R5a: Now available with AMD EPYC 7000 processor for 10% lower cost
R5.metal Bare Metal instances
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
X1 and X1e: Large-scale memory-optimized
For memory-intensive workloads and very largein-memory workloads
32:1 GiB to vCPU ratio
High-performance databases, Large in-memory databases (e.g. SAP HANA), and DB workloads with vCPU based licensing (Oracle, SAP)
For large in-memory workloads
16:1 GiB to vCPU ratio
In-memory databases (e.g., SAP HANA), big data processing engines (Apache Spark, Presto), in-memory analytics
…6 sizes
X1 X1e
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
High Memory instances: Certified for SAP HANA
R3 R4 R5 X1 X1 X1e
Up to 12TB Memory; SAP-Certified
Custom Intel® Xeon® Scalable processor
Native to AWS; Out-of-Box Integration
Simple Management: AWS CLI, Console, IAM
Flexibility to Scale; Resize in Minutes
18 and 24 TB instance coming in 2019
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
I3en: I/O optimized instances
Up to 60 TB NVMe SSD instance storage
~50% lower cost per GB over I3 instances
Up to 100 Gbps of networking bandwidth
Databases, distributed file systems, and data warehousing
Powered by Intel® Xeon® Scalable (Skylake) processors
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
D2 and H1: Dense storage workloads
Lowest cost per storage ($/GB)
Supports high sequential disk throughput
More vCPUs and memory per terabyte of disk
Lower costs for big data uses cases
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Compute-intensive workloads
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
C5: Compute-optimized instances based on Intel Skylake
Custom 3.0 GHz Intel Xeon Scalable Processors (Skylake)
Up to 72 vCPUs and 144 GiB of memory (2:1 Memory:vCPU ratio)
25 Gbps network bandwidth
Support for Intel AVX-512
C5d with local NVMe-based SSD storage
25% price/performance improvement over C4
C4 C5
“We saw significant performance improvement on Amazon EC2 C5, with up to a 140% performance improvement in industry standard CPU benchmarks over C4.”
“We are eager to migrate onto the AVX-512 enabled c5.18xlarge instance size…We expect to decrease the processing time of some of our key workloads by more than 30%.”
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
C5n: Fastest networking in the cloud
33% Increased memory footprint over C5 instances
25 Gbps peak bandwidth on smaller instance sizes
100 Gbps network bandwidth on largest instance sizes
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
z1d: High frequency for specialized workloads
High Frequency instances with custom Intel® Xeon® Scalable Processors running at sustained 4 GHz all core turbo
8:1 GiB to vCPU ratio
Up to 25 Gbps network bandwidth and up to 1.8 TB of local NVMe storage
…6 sizes…6 sizes
z1d.metal Bare Metal instances
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Accelerated Computing WorkloadsApplications that benefit from GPU and FPGA Acceleration
Natural Language Processing
Image and Video recognition
Autonomous vehicle systems
Recommendation Systems
Computational Fluid Dynamics
Financial and Data Analytics
Genomics Computational Chemistry
Virtual Graphic Workstation
3D Modeling & Rendering
Video Encoding AR/VR
Machine Learning/AI High Performance Computing Graphics
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
CPUs vs GPUs vs FPGAs – Architectural Comparison
DRAM
ControlALU
ALU
Cache
DRAM
ALU
ALUControl
ALU
ALU
Cache
DRAM
ALU
ALU
ControlALU
ALU
Cache
DRAM
ALU
ALUControl
ALU
ALU
Cache
DRAM
ALU
ALU
Accelerated Computing Workloads such as training of machine learning models, running 3D fluid dynamics simulations, genomic sequencing and video encoding can take advantage of
parallel compute architecture of GPUs and FPGAs
• 10s-100s of processing cores
• Pre-defined instruction set & datapath widths
• Optimized for general-purpose computing
CPU
• 1,000s of processing cores• Pre-defined instruction set
and datapath widths• Highly effective at parallel
execution
• Millions of programmable digital logic cells
• No predefined instruction set or datapath widths
• Hardware timed parallel execution
GPU FPGA
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
P3 instances: GPU Compute Ideal for workloads needing massive parallel processing power
Training Machine Learning Model
Running HPC Simulations
Rendering 3D models
Video encoding
P3.2xlarge
1 V100
GPU
8 vCPU
61 GB
Mem
P3.8xlarge
4 V100
GPU
32 vCPU
244 GB
Mem
P3.16xlarge
8 V100
GPU
64 vCPU
488 GB
Mem
Up to eight NVIDIA Tesla V100 GPUs
1 PetaFLOPs of computational performance—Up to 14x better than P2
300 GB/s GPU-to-GPU communication (NVLink)—9X better than P2
Support all ML frameworks and model types
P3dn - Most powerful GPU instance in the cloud
Efficiently scale ML model training and HPC simulations across multiple instances with 100Gbps of networking throughput
Fast access to training or simulation data via Amazon S3, network attached file systems or local instance storage
Train larger ML models or process more data via latest NVIDIA V100 GPU with 32GB of GPU memory
Optimize pre-processing of data with 96 vCPU using AWS Custom Skylake CPUs and 768GB of System Memory
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
G4 instances
Designed for machine learning inferencing, video transcoding, remote graphics workstation and other demanding graphics applications
Up to 8 NVIDIA T4 Tensor Core GPUs
AWS-custom Intel CPUs (4 to 96 vCPUs)
Up to 384 GiB of memory
Up to 1.8 TB of fast, local NVMe storage
Up to 100 Gbps networking
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Elastic Inference
Easily add graphics acceleration to your EC2
instance
Configure right amount of graphics
acceleration for your workload
Accelerate application for fraction of cost of
standalone graphics instances
Elastic Graphics
Reduce deep learning inference costs by up to
75%
Easily attach fractional sizes of a full GPU
instance to EC2 or SageMaker instances
Scale inference acceleration up or down as
needed with EC2 Auto Scaling
Choice of accelerators for specialized workloads
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
2. Use Pre-built Accelerators from F1 Partners
F1 Instances: First Cloud Instance with FPGA Accelerators
Speed up applications over 30x using hardware acceleration• Genomics sequencing
• Financial computing
• Engineering simulations
• Image and video processing
• Big data and ML
• Security, compression
New – f1.4xlarge size to optimize price/performance
F 1 i n s t a n c eW i t h y o u r c u s t o m l o g i c
D e v e l o p , s i m u l a t e , d e b u g& c o m p i l e y o u r c o d e
Package as FPGA Images
1. Develop Custom Accelerators or
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Broadest and deepest platform choice
Categories Capabilities Options
Choice of processor(AWS, Intel, AMD)
Fast processors(up to 4.0 GHz)
High memory footprint(up to 12 TiB)
Instance storage(HDD and NVMe)
Networking(up to 100 Gbps)
Accelerated computing(GPUs and FPGA)
Bare Metal
Size (Nano to 32xlarge)
Amazon Elastic Block Store
Elastic Inference
Elastic Graphics+ + =
Compute intensive
Memory intensive
Burstable
Storage (High I/O)
Dense storage
GPU compute
Graphics intensive
General purpose
200+instance typesfor virtually every workload and business need
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon EC2
Resources
Instances
Storage
Networking
Availability
Regions and AZs
Outposts
Management
Deployment
Monitoring
Administration
Purchase Options
On Demand
Reserved
Spot
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon EC2 instance store
Local to instance
Non-persistent data store
Data not replicated (by default)
No snapshot support
SSD or HDD
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon Elastic Block Store (EBS)
EC2 instance
EBS volume
EBS Snapshot
gp2 io1
st1 sc1
EBS SSD-backed volumes
EBS HDD-backed volumes
Amazon S3
Block storage as a service
Create, attach, modify through an API
Select storage and compute based on your workload
Detach and attach between instances
Choice of magnetic and SSD-based volume types
Supports Snapshots: Point-in-time backup of modified volume blocks
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon EC2
Resources
Instances
Storage
Networking
Availability
Regions and AZs
Outposts
Management
Deployment
Monitoring
Administration
Purchase Options
On Demand
Reserved
Spot
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Flow Logs
Amazon Virtual Private Cloud (VPC)
Virtual Private Cloud
Provis ion a logica l ly i solatedcloud where you can launch
AWS resources into a v i rtual network
VPC EndpointsPrivate and secure connectivity to Amazon S3 and Amazon DynamoDB
Security Groups & ACLs
NAT Gateway
Amazon S3 Amazon DynamoDB
Shared VPC allows multiple accounts to launch their applications into a VPC
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Simplifying the network with AWS Transit Gateway
Transitgateway
Amazon VPC
Amazon VPC
Amazon VPC
Amazon VPC
AWS directconnect gateway
Customergateway
VPNconnection
Network topology today After Transit Gateway
VPN connectionCustomer gateway
Amazon VPC Amazon VPC
AWS DirectConnect Gateway
VPC peering
VPC peering VPC peering
Amazon VPC Amazon VPCVPC peering
VPN connection
VPN connection
VPC peering
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon EC2 Resources recap
EBS EBS EBS
VPCVPC
EBS EBS EBS
EBS Snapshots
Amazon S3
Instance
Virtual Machine Configuration
Running or Stopped VM
AMI
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon EC2
Resources
Instances
Storage
Networking
Availability
Regions and AZs
Outposts
Management
Deployment
Monitoring
Administration
Purchase Options
On Demand
Reserved
Spot
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
AWS global infrastructure
SLA of 99.99% availability
21 geographic regionsA region is a physical location in the world where we have multiple Availability Zones
64 Availability Zones Distinct locations that are engineered to be insulated from failures in other Availability Zones
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
AWS Outposts: Extending the CloudSimple extension of AWS Region
AWS Region
(Oregon) Developer
AWS Outposts
AWS Outposts Low latency apps
AWS designed and fully managed infrastructure
Deployed in any datacenter or edge location
Runs key AWS services locally
Native access to full AWS services platform
Modern orchestration and deployment patterns
Seamless extension of AWS Region
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
AWS Outposts
Fully featured AWS services delivered by regional AWS control plane
Same Programming Interface —Standard AWS SDK and CLI
Same Functionality — Leverage full functionality of AWS services
Same Deployment Pipeline — Use the same deployment systems as in the cloud
Same Monitoring and Automation –Use the same metrics, reporting, and operational tools
Low latency apps need high fidelity cloud experience on-premises
Low Latency Telco Operations
Legacy Applications
Financial Services
Content Production, Distribution & Gaming
Real-time Inference
Industrial Automation
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon EC2
Resources
Instances
Storage
Networking
Availability
Regions and AZs
Outposts
Management
Deployment
Monitoring
Administration
Purchase Options
On Demand
Reserved
Spot
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Launching instances
Tags
Launch parameters
User data
Network interface
Placement
AMI ID
EBS volume
Instance type
Launch
Block device mapping
Console
CLI
API
Instances
Templatize launch requests in order to streamline and simplify future launches
Consistent experience
Simple Permissions
Governance and Best Practices
Increased Productivity
with Launch Templates
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Cluster Spread
Placement Groups enable you to influence our selection of capacity for member instances, optimizing the experience for a workload
EC2 places instances closely together in order to optimize the performance of inter-instance network traffic
EC2 places instances on distinct hardware in order to help reduce
correlated failures
Placement Groups
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Fleet managementReplace unhealthy instances
Dynamic scalingScale to demand
Amazon EC2 Auto scalingDynamically react to changing demand, optimize cost
EC2 instances EC2 instances
ELB ELBCPUutilization
Auto scaling group Auto scaling group
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
On-premises capacity provisioning
Time
Load
/Cap
acit
y
Capacity provisioning with target tracking
Time
Load
/Cap
acit
y
Capacity provisioning with predictive scaling and target tracking
Time
Load
/Cap
acit
y
Ramp capacity before you need it with Predictive Scaling
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
AWS Systems Manager: Operate Safely at Scale
Cloud
and
On Premises
Linux
and
Windows
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
AWS Resource Access ManagerSecurely share AWS resources with other accounts or AWS organizations
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon EC2
Resources
Instances
Storage
Networking
Availability
Regions and AZs
Outposts
Management
Deployment
Monitoring
Administration
Purchase Options
On Demand
Reserved
Spot
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon EC2 purchase options
To optimize EC2, combine all three purchase options!
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon EC2 Reserved Instances pricing
Discount up to 75% off of the On-Demand price
Steady state and committed usage
1- and 3-year terms
Payment flexibility with 3 upfront payment options (all,
partial, none)
Reserve capacity or opt for flexibility across AZs and
instance sizes
Convertible RIsChange instance family, OS,
tenancy, and payment
1-Year Convertible RIs
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon EC2 Spot pricing
Increase throughput up to 10x while staying in budget
Launch through AWS services (ex. Amazon ECS, Amazon EKS, AWS
Batch, Amazon EMR) or integrated third-parties
Or containerized workloads
Lean on Spot for these workloads!
Spare Amazon EC2 capacity at savings of up to 90% over On Demand
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon EC2 FleetA single API that optimizes the provisioning of capacity across different instance types, AZs, and purchase options
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
To optimize Amazon EC2, combine purchase options
for fault-tolerant,
flexible, stateless workloads
Now with Hibernate for Spot and On-demand
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Amazon EC2
Resources
Instances
Storage
Networking
Availability
Regions and AZs
Outposts
Management
Deployment
Monitoring
Administration
Purchase Options
On Demand
Reserved
Spot
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I T
Thank you!
S U M M I T © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Raj Pai
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.S U M M I TS U M M I T © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.