6. eBay Private Cloud
Our Cloud Scale
6
IaaS PaaS Frameworks Monitoring
4 PB
Managed Storage
Managed VMs
2 1 0 0 0 0,
Managed BMs
5 9 0 0 0,
Deployments per Month
5 0 0 0 0 0,
Pools
1 2 0 0 0,
Pools
2 7 0 0,
8 8 B
Hits per Day
VMs
6 0 0 0 0,
4 M
Metrics per Second
7 B
URL Logs per Day
7 0 0 TB
Logs per Day
Applications
4 0 0 0,
7. Challenges and Learning
From OpenStack to a Cloud Service able to support our needs
7
Network
Design
Net/Info
Sec
Infra. On
boarding
CI
Builds
Config.
Mgmt.
High
Availability
Cloud
Monitoring
Metrics
Log
Processing
Alerting
Incident
Resolution
Customer
Support
SLA
Upgrades
Capacity
Planning
Remediation
Auto-
Scaling
Scale Out
Metering
and
Chargeback
Monitoring
and
Alerting
CLOUD IS
A
SERVICE!
User
Experience
OpenStack
APIs
OpenStack native
components
Required eBay
developments
8. Operating Principles
8
Customer oriented Open source first
Simplify,
simplify, simplify
Automate everything,
minimize manual work
Obsess over
operational metrics
Engineered systems
with automation
patterns – no one offs
9. Lessons Learned
9
Hard part is turning OpenStack into
a Cloud with ecosystem of tools
and processes
Upgrades
Debugging
Tech Difficulties
OpenStack for an enterprise
requires long term journey and
committment
Infrastructure programmability can
hold you back
Reliability at scale is low without a
lot of work
Complexities
Ramp up time is high
Requires top notch engineers who
are tuned into open source and
community
More engineers & dev. ops, less
tradtional ops teams
Personnel
10. Future Directions
10
Consistent create-build-
ship-run experience
Enable Containers, Micro-
Services
Scalable XaaS interfaces
Declarative intent
Lights Out Management:
auto everything
Efficient scheduling
Enforce policy on each
interaction of an app
Adaptive to dynamic app
changes
Distributed policy
enforcement points
Self contained and
independent
Fully automated
infrastructure (last mile)
Software defined networks
Enable Cloud
Native Stack
Enable Cluster
Management
Programmable
Application
Security
Enable SW
Defined
Data Center
Agile & Efficient Available & Efficient Secure & Efficient Available & Efficient
Cloud effort at eBay has been about Agility rather than Operational Efficiency. In the Agility part, being able to drive innovation through developer agility has been key. For several years we have been driving towards empowering our developers and making things fully self service
Being able to move fast is a key differentiator. Companies that don’t innovate get run-over.
Cloud is an enabler for agility and that should be the primary driver
Top line revenue growth is driven by innovation
Innovation driven by Developer Agility
Our primary driver was Agility
Good news is that it is not orthogonal. Being more agile needs more automation which automatically leads to higher OE
That said, we have actually focused on OE separately
Provisioning form 4 weeks (160 hours +4 w x 5 days x 8 hours) to 30 mins (0.5 hours)
App deployment from 4 h to 10 mins (.2 hours)
What does it take to turn OpenStack into a Cloud Service on top of OpenStack