Best of Breed OpenStack Compute & Block Storage Cloud... .pdf
1. Deploying a best of breed OpenStack
Compute & Block Storage Cloud
…with ass-kicking VMs to show for it
Adam Carter David Medberry John Griffith
Director of Product Management Cloud Engineer PTL Cinder
adam@solidfire.com david.medberry@canonical.com john.griffith@solidfire.com
2. Agenda • What did we set out to accomplish
• How did we get there?
– Compute (Nova) environment
– Block Storage (Cinder) environment
– Deployment via Ubuntu Charms
• What do we have to show for it?
• What we learned along the way
• Where to from here
• Q&A
3. What did we • A blueprint
• A reference architecture
set out to – For any OpenStack deployer looking to stand
accomplish? up a production-ready compute (Nova) and
block storage (Cinder) environment
• For use cases such as
– IaaS
– DBaaS
• Emphasizing the attributes of
– Predictable performance
– Quality-of-Service
– Ease of Use
4. OpenStack Compute Environment
• Folsom on Ubuntu 12.04 via the Ubuntu Cloud
archive
– http://ubuntu-cloud.archive.canonical.com/
• Multi-node nova-compute and distinct nova cloud
controller
• Messaging node with Horizon, Keystone, mysql,
and Rabbitmq
• Swift
5. Deployment
• Juju/Charms
• Applying a generic load to a 'local' volume is not an
ideal Juju use case, but it can be done
• Customizing an image may solve the same problem
more efficiently if network bandwidth is an issue by
pre-loading all packages and applications
• And co-locating an Ubuntu mirror in your cloud is
always a win (aka our colo bandwidth was not
designed for so many instances)
6. OpenStack Block Storage Environment
• Cinder Block Storage Service
• Folsom version and drivers straight from Ubuntu
packages
• Volume provisioning and iSCSI CHAP via
SolidFire OpenStack driver
• Control QoS attributes outside of OpenStack into
the SolidFire API (today)
8. Where to from • Blueprint for Grizzly Cinder (this week)
here • Finish the reference architecture and
publish it
• Cinder development in Grizzly
9. What we learned along the way…
• Setting up an OpenStack cluster still challenging…even for
people with significant experience with Essex – things
have moved, options have changed
• Critical to know the key scaling factors and your use model
• Examine all the nova and cinder defaults and adjust for
your use case
• Be prepared if you update from Essex to Folsom (or
Folsom to Grizzly) to encounter config changes you may
not know about
• Be ready to fix bugs as you go with a new release (we
found Folsom, SolidFire, and Juju bugs as we went)