OpenStack Juno: Continuing Enterprise Maturity

The official release of Juno is coming closer and the program teams are working hard to provide a number of improvements in addition to new features and capabilities in the various OpenStack services. Guillaume Aubuchon from DigitalFilm Tree summed up the entire work saying,

“Icehouse was very enterprise-oriented and Juno is the maturation of that enterprise oriented OpenStack”.

Aubuchon isn’t the only one who finds Juno promising. Das Kamhout, Cloud Orchestration Architect, Principal Engineer at Intel, says that his organization is also following the Juno release closely. There are a number of different groups at Intel using OpenStack, but the two largest deployments are in Intel IT, one deployment being for general purpose usage for all Intel employees and the second deployment focused on exposing Intel’s physical infrastructure in their HW labs.

Like many enterprises, however, Intel is on an older version of OpenStack; they are currently in the process of rolling out Havana in support of their Single Control Plane (1) but there are some new features in Juno that have them excited and wanting to upgrade faster.

“We bucket just about everything into four categories; lowers our TCO, increases performance, increases resilience, and improves end user experience and therefore productivity. We have a goal to have an extremely feature rich Private Cloud which is part of our Hybrid Cloud implementations, and we expect to run this at low TCO. So just about all of this comes into our thinking about what is important for us.”

The advances in replication and exposing cluster backends in Trove DBaaS (database as a service) are exciting to Kamhout. In addition, getting LBaas (load balancing as a service) to support more use cases such as layer 7 switching, HA and SSL termination are key to getting Intel to move away from their internal code to a community implementation.

DigitalFilm Tree uses OpenStack for scalable storage on multiple TV shows and feature films in addition to scalable compute and some aspects of network integration as well. When it comes to the Juno release, there were a few things that stood out for them.

One of these was the degree of network virtualization being added. According to Aubuchon, better integration with their Juniper stack as well as better integration with firewall technology will enable them to run a firewall on top of bare metal and that’s a huge advantage for DigitalFilm Tree.

Aubuchon also likes the Docker integration. “We’re running more and more in containers and even subsets of virtual machines. Containerization is really big for us; the ability to run one transcode process in a Docker container in our OpenStack is huge.”

Another area that stood out is the Swift object storage. Aubuchon notes that before OpenStack they had to transcode material and then deliver it. With OpenStack they can run a containerized process like a video transcode within that storage, even transcode on the fly. “Containers and objects are a wonderful combination.”

When considering TCO, Kamhout says the advancement in Swift on storage policies will provide more flexibility and allow them to tune their costs closer to the SLAs they have to meet. “We are mostly a Ceph shop today, but the advances in Swift on storage policies will allow us to save on the 3 copies for everything approach we have been using.”

Shilla Saebi is on the Cloud Operational Support team at Comcast and provides operational infrastructure support for a wide-ranging collection of Comcast’s customer-facing national product offerings. For her, official integration of Sahara is an important advancement for OpenStack.

“Data processing is an important feature, in my opinion. Bringing Big Data and OpenStack together is key.”

The Keystone to Keystone federation is also making great process says Kamhout. “We ran a POC earlier this year which we will publish the results on, but it fell short of getting real federation between two distinct keystone identity stores, so we really want to see this come through in Juno.”

Kamhout sees some of the standardization efforts happening between cold migrate, live migrate, resize/rebuild is promising, as well as Evacuate Host exposure in Horizon although they still want to see more in terms of HA for the services.

“There still seems to be a disconnect in the community between providing resilience and forcing everyone to rewrite their apps. We are very pragmatic, and we expect that we can run plenty of old code and have the VMs stay on…. So we still want to see a resilient infrastructure and things like live migration solid.”

“Also for our tenants, getting some of the Alerting API exposed into Horizon is going to help us get more users to switch over to using OpenStack for their basic alerting and that single pane of glass we all dream of.”

Other areas of interest to Kamhout include the new Glance metadata configurations table and the new utilization based scheduling which will give them a more holistic perspective on where workloads are placed.

There are many things in Juno Aubuchon is looking forward to, but there’s something from the Juno release he’s trying out right now – the scheduler. Even though most of his business is run dynamically, Aubuchon says it’s still nice to have a scheduling component, especially with Compute, where he can take things offline, “even if it’s just to save electricity”.

There are still ways DigitalFilm Tree can increase their usage of OpenStack according to Aubuchon, and the Juno release is offering a few things that might make that happen. He said that the bare metal provisioning aspect – the changes to Ironic – are going to increase their use of OpenStack in the field.

“Right now we dynamically provision bare metal at the GFT, but when we have an OpenStack instance on a show that’s sort of a static instance, some of the changes we see are going to allow us to dynamically scale in a geo-co-located fashion. We have this OpenStack that comprises the GFT and what’s on the shows. Now we’re sort of changing and molding those smaller OS instances in the field on demand, just like we do our bigger OS instance.”

Aubuchon also likes the Live Upgrades. He notes that even though they have the unique benefit of being able to refresh and reform on a very short-term basis that doesn’t mean they don’t have a need or necessity for migration. A live upgrade path, Aubuchon says, instead of having to completely rebuild the infrastructure, is a nice addition.

For Saebi, there are many changes in the Juno release that will increase her usage with OpenStack, especially around the community work that she is a part of, but her primary focus is on documentation within the OpenStack Manuals, as well as working to improve the Training Guides, so what’s happening in Juno makes her happy.

Kamhout tracks some of the incubation projects and is excited to see the the virt driver getting added to Nova and the Ironic project moving forward. He also commented that it was encouraging to see a lot of focus on making sure OpenStack is tested more for functional tests and that OpenStack is really getting serious that performance moves forward.

Kamhout’s final thoughts on Juno, “with 3000+ blueprints, and not sure how many are actually going to get in to Juno, it is exciting to see the continued maturity of the platform… now if someone could help us upgrade faster with zero tenant downtime we will be upgrading every day.”

(1) Intel IT deployment handles more than 10% of Intel’s compute environment right now, and as they extend OpenStack this year across their entire existing infrastructure it will start handling almost 100% of Intel’s Traditional IT infrastructure regardless of the backing infrastructure. This is what they call their Single Control Plane – one way for developers and app owners to work with Intel infrastructure, while allowing the infrastructure and HW team to make changes behind the APIs.

Photo by bigpilou // CC NC