OpenStack 2015 – The Year of the Enterprise?

OpenStackSummit Paris 2014This post is the collective work of all the Red Hat Enterprise Linux OpenStack Platform Product Managers who attended the summit.

The 11th Openstack design summit that took place last week for the first time in Europe, brought about 6000 participants of the OpenStack community to Paris to kick off the design for the “Kilo” release.

If 2014 was the year of the “Superuser”, then clearly the year 2015 seems to be about the “Year of the Enterprise“.  The big question is: are we ready for enterprise mass adoption?

More than year ago, at the Openstack Havana design summit, it was clear that although interest in deploying OpenStack was growing, most enterprises were still holding back, mainly due to the lack of maturity of the project. This OpenStack summit, the new cool kid in the Open Cloud infrastructure playground is finally starting to show real maturity signs.

An important indicator for this is the increased number of deployments. The Kilo summit showcased about 16 different large organizations using production workloads on OpenStack, including companies such as BBVA Bank, SAP SE (formerly SAP AG) & BMW.

This huge open-source project now counts 17,141 individual users, 432 supporting companies, 20 million lines of code, 145 organizations contributing code and hundreds of millions of investment dollars as of date, and now has more and more enterprise customers considering implementation of large-scale OpenStack clouds for their production workloads. In fact, according to a Red Hat sponsored IDG survey of 200 enterprise IT decision makers, 84 percent of them plan to deploy OpenStack (compared to last year statistics reporting minimal adoption).

With all this action, with all these vendors and such a larger user base the question remains: is OpenStack ready for the enterprise? Is it stable enough? Is it secure? Can it scale to meet future needs? Can it run enterprise workloads with the level of robustness the market expects?

Coming of age
OpenStack was started four and a half years ago and most of the community members are over the “works on my laptop” phase. The Ops Summit that took place on Thursday, in parallel to the Design Summit, was packed. The increased number of operators that have been deploying OpenStack and have been through several upgrades, is a sign of OpenStack’s coming of age.

This long term use ends up forcing the projects to become more stable and adapt to larger datasets and continuous use. Ceilometer, for example, after getting rid of its last non active/active service, is schedule to get a new time series oriented backend to cope with larger datasets through Gnocchi, a new project that uses Swift, OpenStack’s object storage project, as a backend.

Upgrading deployments to keep them in synch with the project’s rapid pace of innovation is a must and several sessions were dedicated to discussing upgrade strategies to:

  • reduce downtime
  • minimizing manual errors with more automated tests
  • understand how to isolate failure domains
  • improve high availability across the whole system
  • and how to keep compatibility between modules running different versions to increase stability and robustness.

Several projects addressed these problems using different approaches and many sessions were about sharing war stories on how early adopters overcame problems and what was learned in the process.

The Kilo design summit had several sessions on further improving enterprise use cases such as better integration between Keystone and Horizon, hierarchical multi-tenancies and Keystone federations with Single-Sign-On. Horizon developers had tightened discussions with operators and enterprise users and held discussions on making Horizon pluggable and easier to integrate with.

More monitoring, management and orchestration tools, required by enterprise users, were also in the spotlight. Those are crucial components as deployments get bigger and bigger. Deployments with hundreds of nodes aren’t rare anymore, they moved beyond Proof of Concepts and graduated to production. Growing pains aren’t pleasant, but they exist due to OpenStack’s users success.

This success brings an increased number of telecommunication operators who are now starting to test OpenStack to deploy their core functions.

Network Function Virtualization
After the Atlanta summit we reported that OpenStack was seeing an increase in use in the telecommunications industry. For anyone in doubt, during the Paris summit, service providers, equipment providers, vendors and related industry groups were present in large numbers. Telecommunications and NFV use cases presented keynotes, had an entire track, and for the first time there was a NFV targeted design summit session.

There was also an opportunity to convene a Telco Working Group to build on the previous efforts of the OpenStack NFV subteam to continue driving required functionality into OpenStack.

From a telecom operation perspective, in the Telco Ops Summit session we discussed the team objectives and the requirements to focus. In general, those can be categorized to three:

  • Operational Requirements: which are similar to enterprise requirements and required as OpenStack continues to mature. Here we can find requirements like High Availability (HA), easier provisioning and better monitoring.
  • Telecom Requirements: which are derived from telecommunication service providers use-cases. Here we can see an increasing demand for IPv6 networking, data center interconnect (DCI) solutions and focus on large-scale operations.
  • NFV Requirements: which were primarily focused on performance and deterministic placement of VMs on the Compute and Networking parts. As we expand the NFV community and gathering more use-cases, there is a need to extend the discussion here to other OpenStack components to address topics like orchestration and service chaining.

As the NFV ecosystem grows up, we see an increasing number of vendors which are providing their virtualized network functions, or VNFs. VNFs may consist of one or more virtual machines running different software and processes, effectively replacing the custom network hardware devices. While VNFs may be created as elastic cloud applications, some of the existing VNFs available today are built as traditional applications. This requires OpenStack to provide a common ground and support both types of workloads easily. One of the main areas of focus in this domain is High Availability (HA) in the different levels of the infrastructure.

While NFV introduces new requirements for management and orchestration, the main focus so far was on deterministic performance of the infrastructure. On the OpenStack networking service (Neutron), we are expecting improvements to the SR-IOV driver introduced in the previous cycle. There is also some active work to enable DPDK support for enhanced packet processing.

One of the big themes in the NFV sessions was the expressed need to have a better line of communication between the different parties involved, namely the operators, telecommunication service providers and developers. The general consensus was that user stories, use-cases, and more detailed requirements are needed in order to drive the NFV development further.

The summit was also an occasion for Red Hat to make a few announcements:

Session Videos

If you missed the summit, you still have the chance to view the presentations on YouTube.  For your convenience, here is a list of the sessions that were led by the team at Red Hat.

As always, this last summit was a really exciting time for the whole community in general, and for Red Hat as well, as the increased adoption in the enterprise was clearly reflected in the numerous interactions we had during the Summit.  It’s now time to get to work on all that has been discussed. We’re looking forward to the Vancouver Summit in 6 months. We hope to see you there.

 

 

 

 

[1]The use of the word “partnership” does not mean legal partnership between Red Hat and Wipro.