IT pros from large-scale shops gathered online this week to discuss their experiences with the new OpenStack Havana release, as well as what to expect with next year's version, codenamed Icehouse.
Panelists on a Google Hangout organized by the OpenStack Foundation this week praised Havana's autoscaling features, networking updates and support for rolling upgrades, but they said there's still plenty to be desired, from live migration and full-fledged Database as a Service support to federated identity management.
Autoscaling is accomplished through the combination of two projects that left the incubator stage with the Havana release: OpenStack Orchestration, also known as Heat, and OpenStack Metering, also known as Ceilometer. OpenStack Metering monitors the environment for resource constraints, and OpenStack Orchestration spins up new resources as needed.
PayPal runs incubator versions of the projects in its data centers, but the new integration between the two means much more granular ways of adjusting resources according to need, said Anand Palanisamy, architect in the PayPal infrastructure engineering team.
Previously, OpenStack Orchestration could only adjust resources based on CPU usage, Palanisamy said. Now, there's support for more metrics, such as RAM and disk I/O.
OpenStack wish list
Other items remaining on the users' wish lists include performance analysis, testing of patches and updates, and more operational tools for administering ongoing OpenStack deployments.
Orchestration represents a much-needed tool for companies that develop private clouds, according to Guillaume Aubuchon, senior post consultant at DigitalFilm Tree, a media and entertainment IT consulting firm in Los Angeles. He works with Warner Bros. and other major companies on private cloud projects.
"We're also working heavily with the Heat team at Rackspace to move workloads back and forth between private and public clouds" based on demand, Aubuchon said.
For PayPal, the standout features in the Havana release include OpenStack Neutron support for Firewall as a Service and Load Balancer as a Service.
"Being a payments company, we are extremely cognizant of security, because it's a core component of our product and our experience," said Saran Mandair, senior director of infrastructure engineering at PayPal, during a separate interview. In fact, PayPal had developed its own distributed firewall that it will now share with other OpenStack developers, Mandair said.
Top on the wish list for attendees at OpenStack Summit in April was support for automated rolling upgrades, which now appears in this release. Previously, users had to deal with a disruptive manual process that ran the risk of running into circular dependencies between different components of OpenStack.
PayPal also sees another use case for rolling upgrades, including OpenStack updates in its continuous integration and continuous development processes.
Live migration, federated identity management still missing
Boot From Volume, a new feature of the Cinder storage project that sets the stage for supporting live migration of virtual machines, was of interest to Das Kamhout, principal engineer at Intel Corp. who spoke during the Google Hangout.
Initially, when deploying OpenStack in 2011, Intel only provisioned cloud-aware, scale-out apps into the platform, but it's looking to migrate traditional and legacy apps into the platform as well, according to Kamhout.
Boot From Volume, live migration and the ability to evacuate a host will all come in handy to protect those applications that rely on the resiliency of the infrastructure, Kamhout said.
"We need Heat to support live migration as well, which is coming," he said.
"We don't want to have multiple Keystones for every cell we build," said PayPal's Palanisamy.
Database as a Service features are part of an incubator project in this release called Trove, expected to mature with Icehouse next year. PayPal's Mandair said this will be key for developers at PayPal who want to provision and run their own databases for code-testing purposes.
Other projects expected with Icehouse include bare-metal provisioning, codenamed Ironic; a Hadoop implementation called Savannah; and queuing, dubbed Marconi.
"If you need to move traffic to a different cell of the cloud, how do you actually move [all the] traffic?" said Mandair. "If you see an error message, how can you resolve an incident rapidly without affecting either your cloud users or the applications that are residing on that cloud?"