Global Large Scale Cloud Design
Cloud Native Enterprise Infrastructure
Fast Retailing Co., Ltd.
Keito Fukuda
Lead Technical Architect
Fast Retailing Co., Ltd.
Responsibilities:
• Digital Services
• Mobile Services
• O2O
• Backend Platforms
• Infrastructure/Architecture
Kenichiro Ara
Infrastructure & Communication Lead
Fast Retailing Co., Ltd.
Responsibilities:
• Enterprise infrastructure/architecture
• Global WAN
• Office/Store infrastructure
• Corporate infrastructure
• ICT technologies
JPY1,380,000,000,000
2,75389,580 20+
Revenue
StoresEmployees* Countries
As of Aug. 2014
*Including Full-Time Employees, Contract Workers, Part-Timers
4,500
9,000
13,500
18,000
1992 1995 1998 2001 2004 2007 2010 2013 2015
(JPY 100M)
Est.
Group Revenue
1. AWS at FR
2. Cloud Design Strategy
3. Global Network Design
4. Enterprise Cloud Strategy
Agenda
1. AWS at FR
2. Cloud Design Strategy
3. Global Network Design
4. Enterprise Cloud Strategy
Agenda
UNIQLO Recipe Wakeup Clock Calendar Looks
UTme Bicqlo Kids Camera
EC Site UNIQLO APP UTme APP
20,000,000,000
req/month
100,000 45G
req/sec bps
As of May. 2015
100+ 1,300+
systems instances
As of May. 2015
# of instances on AWS
500
1,000
1,500
2,000
2010 2011 2012 2013 2014 2015
0 10 50
300
1,000
2,000
Agenda
1. AWS at FR
2. Cloud Design Strategy
3. Global Network Design
4. Enterprise Cloud Strategy
Independency Maintainability
Availability
Recoverability Replicability
Scalability Visibility
Cloud Design Best Practices
Independency
Make System Loosely Decoupled(Micro-Service)
1300+ Instances
100+ Systems
30+ Vendors
Independency
Many Instances Co-Existing in Cloud
Each System Sharing Recourses and Assets
Independency
A B C D E
Each System Sharing Recourses and Assets
Independency
A B C D E
Infra Cost/Procurement No Longer Problem
Independency
Make Respective System Independent to Minimize Dependency
A B C D E
Independency
Use Security Group per System to Satisfy Security
A B C D E
Maintainability
Take Full Advantage of Full-Managed Services(PaaS)
Maintainability
System Getting More Complicated
Mail
Queue
RDBMS
Cache
Search
NoSQL
BigData
Push DNS
Maintainability
System Getting More Complicated
Mail
Queue
RDBMS
Cache
Search
NoSQL
BigData
Push DNSDo Not Reinvent Wheels and Babysit
Maintainability
Minimize Worrying about Infrastructure
Scalability
ELB + AS Ready Instance with Auto-Scaling
ELB AS Ready EC2 Auto Scale
Scalability
Probably Scalability Not Required at the Beginning
Scalability
Probably Scalability Not Required at the Beginning
System Requirements Always Change
“AS Ready”
Stateless + AutoStart
Scalability
Make System Scalable from the Beginning
Scalability
Scale System When Necessary
Availability
Multi-AZ/Region with DNS Failover
Availability Zone
region
Route53 Region Availability ZoneELB
Availability
Availability Zone A
Probably Availability Not Required at the Beginning
Availability Zone A
Availability
Probably Availability Not Required at the Beginning
System Requirements Always Change
Availability Zone A
Availability
System Designed Assuming Infra Never Goes Down
Availability Zone A
Availability
System Designed Assuming Infra Never Goes Down
Do Not Assume Instance will Never Go Down
Availability Zone A Availability Zone C
Availability
Do Not Retain Availability in Single Instance, Make It Overall
Recoverability
Easy to Recover from Troubles
AS Ready EC2
Recoverability
System Happens to Go Down
Monitoring
TroubleShoot
Recoverability
Developer Begins to Troubleshoot
TroubleShoot
Recoverability
Do Not Make Developers TroubleShootMonitoring
Do Not Take Long Time to Get it Recovered
Restart
Recoverability
Make System Recovery Easier
“AS Ready”
Stateless + AutoStart
Monitoring
Replicability
Easy to Clone a System
CloudFormation Elastic Beanstalk CodeDeploy
Replicability
Production
Only Production Environment Necessary at the Beginning
Production Staging Integration StressTest Development
Replicability
Multi-Environments Required as We Go
Replicability
Production Staging Integration StressTest Development
Make Cloning System Easy
Portability
From IP-based to“Owned”Name-based
Route53
Portability Users
System Needs to Connect with lots of Endpoints
Portability Users
System Needs to Connect with lots of Endpoints
Do Not Get All Endpoints Connected thru IPs
Users
org.xxx.uniqlo.com
read-db.xxx.uniqlo.comdb.xxx.uniqlo.com
cache.xxx.uniqlo.com
Portability
Assign“Owned”Hostname to Endpoints for Maximum Flexibility
static.xxx.uniqlo.com
Users
org.xxx.uniqlo.com
Portability
Easy System Update with Blue-Green Deployment
Visibility
Collect as Much Data as Possible
ConfigCloudWatch CloudTrail
Visibility Users
Modern System Consists of Full-Managed Services
Users
Visibility
Difficult to Figure out Root Cause
Users
Visibility
Difficult to Figure out Root Cause
Not Everything under your Control on Cloud
Monitoring
Visibility
Collect as Much Data as Possible
Do NOT Carry Over
Enterprise Best Practice to Cloud
Agenda
1. AWS at FR
2. Cloud Design Strategy
3. Global Network Design
4. Enterprise Cloud Strategy
Tokyo
Sydney
Singapore
Ireland
Sao Paulo
California
Oregon
Frankfurt
Beijing
Virginia
Multi-Regions for Global Service Distribution
Production
Staging
Development
Management
4 VPCs in Each Region
Region
Production
Staging
Development
Management
B2C/B System Residing Together, Each VPC Completely Isolated
Region
Production
Staging
Development
Management
Management VPC Reachable to All VPCs
Region
Production
Staging
Development
Management
10G x 2
1G x 2DX for Only System Connection
Direct Connect
Production
Staging
Development
Management
10G x 2
1G x 2
Customers
Employees
Developers
Access
Management
Production
Staging
Development
Same VPC Design across Regions
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Cross Region VPC Peering
Management
Production
Staging
Development
Mesh Topology for High Availability
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Cross Region Connection with VyOS
VyOS
VyOS
VyOS
VyOS
VyOS
VyOS
VyOS
VyOS
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
VyOS
VyOS
VyOS
VyOS
VyOS
VyOS
VyOS
VyOS
Difficult to Maintain Configuration
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
VyOS
VyOS
VyOS
VyOS
VyOS
VyOS
VyOS
VyOS
Orchestration for Auto Configuration with Serf
Management
Production
Staging
Development
Access from Consumers
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Users Users
Users
Management
Production
Staging
Development
Access from Developers
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Management
Production
Staging
Development
Developers Developers
Developers
Public(DMZ)
Private
Secure
3 Segments in Each VPC
VPC
Public(DMZ)
Private
Secure
Access Control by NACL
VPC
Public(DMZ)
Private
Secure
Hosting in Appropriate Segment
VPC
Agenda
1. AWS at FR
2. Cloud Design Strategy
3. Global Network Design
4. Enterprise Cloud Strategy
Quick Decision Making for AWS Migration
2014/3:
2014/4-10: Verification & Adaptation
2014/11: Director & CIO Approvals
2014/12: CEO Approval
Current Enterprise AWS Architecture
Data

center
1G x 2
10G x 2
Production
Verify & Dev
Paradigm Shift Required -Mindset
Instance Gets Down
Hardware Crushes
Requirement Changes
WAN-Based Internet-Based
Paradigm Shift Required -Network
Encryption & Rights Management
Real-Time Monitoring & Treatment
Paradigm Shift Required -Security
Fast Retailing IT Team
Direction towards 2020
12,500
25,000
37,500
50,000
1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
(JPY 100M) Group Revenue
Disrupt Traditional Enterprise Systems
In-House Development
Open Source
Now Hiring!
http://www.fastretailing.com/employment/career/jp/fr/it/

AWS Summit 2015 Tokyo Breakout: Global Large Scale Cloud Design and Cloud Native Enterprise Infrastructure at Fast Retailing