BMW is preparing to replace its in-house developed private cloud platform with OpenStack open source system, deciding against proprietary cloud management vendors due to the risk of lock-in.
The car manufacturer started developing its Python-based private cloud management system in 2011, but began seeking alternatives after it found that conventional virtualisation and automation could not reap the efficiencies it required.
Speaking to ComputerworldUK at OpenStack Summit in Paris, BMW’s head of data centre infrastructure, Dr Stefan Lenz, said that deploying a cloud management platform from a commercial vendor would be easier and faster, but “my fear is that then I am too tied to the roadmap of these companies”.
“The problem really is that this [technology] is not yet settled, this is not a stable thing from the point of view of what the software should do. For example, you know what an Excel spreadsheet is supposed to do - it is a very defined task, and it is very mature. OpenStack is still not clearly defined at its borders - such as how storage and networking is handled, and the whole market is still flowing - not only OpenStack.
“If I now choose a commercial vendor I am tied to their roadmap and I need to give them money to buy the software. That money is then an investment that ties me to their roadmap, and I think it is too early to do this. With OpenStack I feel I have more openness, and don’t have to throw it all away and maybe lose the money invested.”
Re-assessing internal private cloud
BMW has four data centres globally, with 98 percent of its servers managed from a Munich operations centre. It runs around 12,000 operating system images, half Linux and half Microsoft, with around 80 percent of its infrastructure virtualised and ‘cloud-ready’.
Three years ago the firm went down a path of “brutal standardisation”, developing a management and orchestration layer to bring all of its automation under one modular framework.
“It was wonderful, it was perfect - we got new efficiencies and everything was great. And then we ran into trouble,” said Lenz. BMW ran into problems running certain applications, for example.
After consulting with staff, Lenz ran a trial of the OpenStack Icehouse release on a single server, before growing this to what he terms a ‘pre-production’ environment, with 16TB of storage and 100 cores.
“We see in OpenStack two major advantages. First we have an API in the data model to describe cloud and virtual instances that will become industry standard. So when anyone develops on that, it will be stable - we do not have to change whole tool chains. That is a big advantage and an area where we failed because we were unable with our internal cloud development to get that stable input.
“The second thing we see is that it is open source and free. No one earns money on the growth of our company, no one comes with unexpected license fees after we have develop a lot of software around it. And we have experienced that a lot. So that is the reason that we believe OpenStack is the right choice at the core of our automation.”
'Prime candidate' for private cloud
OpenStack is now the “prime candidate” to resolve its orchestration layer difficulties, and he believes that a substantial amount of its applications could be running on OpenStack within three years - provided it meets certain criteria such as delivering anticipated cost benefits.
“We will start step by step, and if we find any obstacle, we always keep our conventional [cloud platform] in parallel, so we can move back and forth,” he said.
“We have not made a cost-based decision right now, but if this does not help us to bring cost down by another 10-15 percent - at least - then it is dead.”
One use case could be to improve delivery of resources to develop dev-test teams working on connected car systems, he explained.
“We have a lot of people in the engineering department who do development and testing of software for connected cars, and we have a lot of people who develop on backend systems for all of these things. These people have already asked if they can get on that environment.
“We run about 3,000 applications and we have a lot of people who will be interested to get the test and development instances. So I can imagine that about a third of the 12,000 [operating system] instances go for development and testing.
"That is the potential, if we could put every dev and test on an OpenStack environment there is potential for maybe 4,000 server instances.”
However, he said that, like all large manufacturing firms, BMW has extremely time-critical systems, which means that around 20 percent of its workloads will remain in a conventional operations mode, and “maybe even on physical servers”.
Reaching enterprise maturity
One of the challenges that has faced the OpenStack project has been to move from its origins within IT supplier data centres (and, increasingly, telcos) and into more traditional enterprise settings.
Lenz said that, despite some areas which need improvement, the software is ready for BMW to deploy.
“It definitely has the maturity. It is good software, it has a sound data model and that is the most important thing. It is modular and has the API that is reasonable. So these features look very promising. From the point of view of features of software quality it has a way to go - but so have all of the commercial products,” he said.
“The thing that bothers us most with OpenStack at the moment is the release cycles, and the huge changes we have from release to release,” he said.
“If you are developers then I want to give you one message - to make this industry grade, we need more stability in the future, though it doesn’t prevent from using it now as it is.”
OpenStack - the last cloud survivor?
Although BMW is still in the trialling OpenStack, Lenz said that as long as the project can carry its momentum and not be derailed by a fracturing of the community, the company will continue on this path.
“That ship has sailed,” he explained. “The only thing that could prevent it would be if the people who professionally support OpenStack would not agree on a joint roadmap - for example if someone would now split off their version of OpenStack and go their own way, if OpenStack would diverge into different directions - that would cause OpenStack to die.
“But as long as the guys for OpenStack keep that trunk development stream always stable and reliable, and there are vendors who can handle that an provide services to companies like ours, I think OpenStack will make it and all the other products will disappear.”
He has no plans, however, to partner with commercial cloud management platform vendors.
“My biggest concern now is that I invest now to VMware, Microsoft or any of these and give them money and in three years, just when I go to production they tell that they discontinue the product. And that has happened - not with these companies - but in general,” he explained.
“If you look realistically at CloudStack with only Citrix as a company - and a couple of others - supporting it, how long will they have the power and energy to do so?
“And if you have implemented on that and in three or four years they tell you that they don’t support it any more, what do you do? You have lost all of your investment.”