When the GE Predix Cloud for industrial data launches next January, it will bear little resemblance to the big cloud data centers built by Amazon, Microsoft, and Google.
Microsoft, for example, built a showcase cloud data center outside Chicago suitable for 300,000 servers. "We won't be building any 900,000-square-foot data centers," said Harel Kodesh, VP and CTO of GE's Software unit, supplier of the Predix machine data analysis system and cloud service.
GE's Predix Cloud will require a more decentralized strategy, putting enough centers into North America and around the world to capture the data from the world's largest manufacturing and production systems. Reliable, high speed communication lines from its sites to those of its customers' will be as vital as any other data center element. GE expects its Predix Cloud to collect twice the amount of data already generated by the consumer Internet.
It also realizes that many big industries and manufacturers do not wish to simply turn their data over to GE's IT staff or a GE data centers. Whoever is capturing and analyzing the data will also have to be its custodian, and competitors don't want GE to assume that role.
[Want to learn more about GE's Predix Cloud announcement? See GE Predix Cloud: Industrial Support For Machine Data.]
"The information won't pass through GE's IT fabric. People don't want to feel that GE will have a chance to look at their data," said Kodesh in an interview the day after GE's Aug. 5 Predix Cloud announcement.
Instead, Kodesh will supervise the establishment of the Predix Cloud on neutral, co-location sites, many in existing Equinix data centers. GE will run its own hardware and its Predix Cloud service in the co-lo facilities. It wasn't made explicit, but co-location sites are also the likely facilities where customers' data will be stored.
Early users of GE Software's Predix analytics system are already logging into Amazon Web Services, where the service is currently being hosted. Kodesh said that practice is likely to continue for developers and data managers producing Predix applications. But as they are ready to stage and then launch their production systems, "they'll move their code over into our Predix Cloud centers" in the co-location facilities, he predicted.
But GE is also forced to recognize that if its Predix service is going to be adopted worldwide, it will need to recognize the data sovereignty requirements in Germany, the UK, the Netherlands, and France, as well as many other parts of the world.
"The plan for next year is getting shops first in the U.S., where early development projects will start. But we're following the machine data," and that means establishing the Predix Cloud in Europe, Asia and other parts of the world as well. In many cases, for Predix Cloud to succeed, it must keep the data it collects within the national borders of the country where it's collected.
In the Middle East, Russia, Asia and Europe, as well as the U.S., basic engines of the economy "are considered critical national infrastructure. That's why the data has to stay there," Kodesh said. And GE will build smaller-footprint datacenters inside co-location facilities in each region, instead of relying on a few massive data centers that might require businesses to export their data across national borders.
Kodesh conceded there was little to stop Amazon Web Services, IBM, or Microsoft from trying to offer their own machine data services on their own networks of cloud data centers. But cloud infrastructure might be the lesser part of the story, he said.
GE's Predix system is built on top of GE's experience in manufacturing turbines, locomotives and jet engines. It knows something about what data is crucial to collect, how to analyze it, and what to look for among anomalies, said Kodesh.
"You have to know what it means for there to be a high temperature on half of the turbine blade… You need the ability to know the physics of those models" that get embedded into Predix data analysis applications," he said.
Among other things, such a system needs to recognize machine data from the network edge, such as noisy drill data, clean it if necessary, transform and store it, and then analyze it. Data from sensors on machines may be polluted with external noise, and a data service will have to sort the good data from the bad. The Predix Cloud will have at least 100 micro data services capable of helping customers capture and store their machine data.
"We don't see many could service providers who are manufacturing these industrial machines," said Kodesh. "We humbly believe we're the only ones with the experience" to capture and make effective use of the machine data.
Kodesh conceded that an Amazon or Microsoft may operate large-scale compute infrastructure more cheaply than GE can, but GE will be able to provide high-speed, private communication lines for regulated industries through the extensive interconnection facilities found in the "meet me" rooms of Equinix data centers.
"We're not opposed to working with these providers [Google, Amazon, IBM, etc.] if they can show the right amount of badges and certificates from the local government where we plan to operate," said Kodesh. For starters, though, GE will stick with its plan to operate out of select co-location facilities and Equinix centers.
GE Software teams will be further explaining the Predix Cloud at its industrial data show, Minds & Machines, in San Francisco, Sept. 29-Oct. 1. Afterwards, it will take Predix Cloud out of private beta of -- where select customers such as Pitney Bowes are using it -- to a public beta, during which a longer list of tryouts may occur. In January, Predix Cloud will become generally available.
Although GE consistently refers to the industrial scale of Predix, Kodesh doesn't rule out smaller users. The service will be useful to a wide range of machine data generating businesses, "from the airline with 2,000 jet engines to the company with 10 vending machines at the airport," he said.