4. #rackstackatl
Agenda
• Up first: A tale of two workloads
• Fictitious customer use case: how we’ll structure our discussion
• Proposed solution: looking at OpenStack and VMware together
• Q&A
14. #rackstackatl
• Let the application handle its own resiliency
• Build loosely-coupled distributed systems
• Use the scale out approach instead of the scale-up approach (scale horizontally
not vertically)
• Choose cattle over pets
This translates into certain design guidelines
16. #rackstackatl
Our customer is Acme Corporation
• Wile E. Coyote is a huge customer, and
business is booming (no pun intended)
• Acme runs most of their apps on vSphere,
including some Oracle DBs
• IT recently tasked with building out environment
for new mobile applications
– Been using AWS for prototyping
– Want to bring that in-house but maintain “AWS-like”
experience
– Plan to leverage cloud-native application
architectures
21. #rackstackatl
• Leverage VMware NSX with OpenStack
Neutron for cross-hypervisor networking
services
• Acme developers can spin up their own
isolated logical networks for
development and prototyping
• Production workloads can leverage
existing IP address space for seamless
integration into the rest of the network
yet remain isolated from each other
Networking
Physical Network
(131.107.x.x/16)
Logical Network
(131.107.1.x/24)
Logical Network
(131.107.2.x/24)
Logical Network
(192.168.1.x/24)
Logical Network
(192.168.1.x/24)
23. #rackstackatl
• Must build and maintain multiple Glance images
– One image needed for each hypervisor in the OpenStack environment
– Image metadata associates the image with the appropriate hypervisor
• Development/testing on KVM but production on vSphere can introduce issues
• Networking could be challenging without VMware NSX (or other cross-platform
Neutron-capable solution)
• Organizational and staff readiness to support multiple hypervisors (“Layer 8”
concerns)
Operational challenges of this solution
25. #rackstackatl
Key takeaways to remember
• OpenStack and VMware’s products (including vSphere) are complementary in
many ways
• Workload dictates architecture; different requirements result in different
architectures
• Use the right tool for the job