Prepared by Tim Bell, Ryan Lane and JC Martin, your User Committee. See also: Infographic.
Introduction
In preparation for the OpenStack summit in Hong Kong in November 2013, members of the OpenStack foundation were asked to provide their feedback via a user survey. The goals were
- Profile the user community across geographies and industries
- Understand the current deployments of OpenStack
- Receive input on priorities for the technical and management boards
The format was
- Information about the person concerned
- Feedback on priorities and improvements
- For those with OpenStack deployments, questions on deployment sizes, technologies used in multiple choice format
The survey is open for any one to input their ideas and deployments at https://www.openstack.org/user-survey but a specific campaign was run during September/October to get the latest details.
The previous survey was performed in April 2013 and presented at the summit (see http://www.openstack.org/summit/portland-2013/session-videos/presentation/openstack-user-committee-update-and-survey-results)
As with all surveys, there are risks of errors in that those who report to an anonymous survey do not necessarily reflect the installed user base. For an open source project, this is especially difficult as there is no tracking of deployments.
This report covers the statistics gathered from the survey based on the presentation prepared by J.C. Martin from the user committee with input from Ryan Lane, Tim Bell and Tom Fifield. Further analysis is ongoing for the comments and feedback which will be published later.
Changes since the last survey
The September survey added several new questions and modified some of the originals, with the aim of clarifying some of the responses from the previous survey and gaining deeper insights into the community’s methodologies for OpenStack deployment. To gather the best possible data after the changes, a communications campaign was enacted to encourage previous contributors to update their responses.
A summary of changes is as follows:
- Industry list updated based on “Other” responses from previous survey
- Information sources – removed forums, added ask, planet, ‘other’
- Added “Community Cloud” to cloud types to be in line with NIST definition
- Replaced references to Quantum, and added Orchestration and Metering projects
- Attempted to clarify “User Group” participation, based on poor previous responses
- Better recognition of Continuous Deployments
- Fixed lists of Hypervisors, Block Storage and Network drivers based on current support
- Changed “Number of Users” question from free-form to a pick-list
- Changed “Workloads” question to a select list, based on collating options from previous free-text response
- Added a new question “What do you like most about OpenStack?”
- Added a new question “If you are using nova-network and not OpenStack Networking (Neutron), what would allow you to migrate?”
- Added a new question “What is the main Operating System you are using to run your OpenStack cloud?”
- Added a new question “What tools are you using to deploy/configure your cluster?”
- Added a new question “What are your business drivers for using OpenStack? ”
User Profiles
822 people from 539 different companies responded to the survey, 216 of these were already members of OpenStack user groups which is an encouraging sign of involvement in the community.
Geographically, the community is widely spread with the US responses now being the minority. Given have that this summit is based for the first time outside of America, it demonstrates the global reach of the OpenStack community and the importance to continue with a global approach.
The survey received nearly twice as many answers as the previous round (822 compared to 414) and 387 deployments compared to 187. The national distributions have adjusted a little with the US response share dropping to 38% from 42% with corresponding increases elsewhere.
The industries are clearly dominated by IT companies along with Academic and Telecoms with 80% of deployments. Government, Film/Media and Manufacturing are more limited but there is a trend towards diversification as the previous survey had 85% of deployments in IT/Academic/Telecoms.
Organisation sizes are similar to the previous survey with well spread mixture of small companies to large.
Business drivers were similar across deployments with the emphasis on agility. Nearly half the organisations felt that implementing OpenStack was an effective way to attract talent.
For information sources about OpenStack, there is now a rise in the formal documentation usage such as docs.openstack.org and the operations guide. This reflects well on the efforts that have been placed in this area as it was one of the items highlighted in the previous survey as an area to improve.
ask.openstack.org was also started recently and is now rising up the information sources reflecting the benefits of developing standard Q&A high quality answers.
With Grizzly coming out, there has been a clear migration from Folsom and Essex to Grizzly. Installations on Havana have now started and the sites on trunk have continued to follow that approach.
Private cloud deployments are the majority as in the past survey.
For features, with ceilometer and heat becoming standard components and maturing rapidly, their adoption is accelerating. Bare metal and database-as-a-service deployments are starting to appear.
Over 165 deployments are now in production. This is around double the number in the previous survey (84). Equally, the Dev/QA and Proof of Concept deployments have doubled in the past six months.
Most features follow similar ratios to the previous survey (but deployments are around twice). OCCI was asked for the first time but the usage currently is not widespread compared to EC2 which is enabled over 30% of the OpenStack installations.
For the implementation choices, OpenStack provides many alternatives.
In the storage area, LVM is the largest single deployment technology which probably reflects on the ease of installation. Ceph, however, is available in nearly 20% of deployments. The huge list of storage options illustrates the different configuration choices that sites are making while deploying cinder, especially if there is a deployment for other purposes at the site. Since multiple options could be selected, this could also indicate that sites are trying several different backend storage solutions.
For deployment tools, Puppet comes out on top. However, it is encouraging to see that all but one site considered a deployment tool to simplify the installation and configuration of OpenStack.
Within the different deployments, there is a variety of scale. Many of the proof of concept instances have a small number of virtual machine instances but there are now over 30 clouds with over 1,000 instances, 15 with over 5,000 cores and 11 with more than 1,000 hypervisors. Storage, networking and objects follow similar curves with the smaller instances providing many small configurations and several at large scale.
OpenStack is used for both public and private cloud deployments. The number of responses on these points is significantly less than the total deployments, illustrating that these questions may also be considered sensitive by the deployers.
The following statistics were all gathered by dropping the proof-of-concept reports and focus on the production and dev/qa instances.
The deployment tool space has a number of common solutions. devstack is used on many of the smaller instances, presumably as part of the deployment of test clusters. However, as the number of nodes increases, tools such as Puppet and Chef takeover.
KVM continues to be the most popular hypervisor for production deployments but the variety continues to expand, even including container technologies such as lxc and OpenVZ.
Ubuntu remains the most popular O/S for OpenStack deployments, especially for the smaller configurations.
For network drivers, there are a variety of drivers with the open source vSwitch leading the pack.
Additional References
