This year GE hosted their first annual—and according to them, the world’s first—developer-focused industrial internet conference, Predix Transform. Touted as both an industrial IoT platform and the operating system for the industrial internet, Predix originated in a 2009 realization that while GE heavy equipment generated a tremendous amount of sensor data, the company and its customers had no way to make use of that data. Taken together with GE’s exit from the financial services market (they shed the $42 billion GE Capital unit that same year), Predix represents a fundamental shift in GE’s long term strategy.
A handful of large companies have made successful pivots in recent memory; Adobe’s switch from shrink-wrap software licenses to cloud subscription revenue is one I particularly admire. However, GE is attempting a more fundamental, and more massive, shift. Known mainly for physical and (usually) enormous manufactured goods, this 120-year-old company wants to use Predix as the platform to transition to become a top 10 software company by 2020.
Transform offered the chance to explore some larger questions: how exactly, would GE execute this transformation? Would their partners and customers see this as an opportunity and come along for the ride? Transform also offered a chance to find out what real developers were saying about the Predix platform. I came away from the conference with a strong consensus from developers: while growing pains abound, the company is making a strong, consistent, even urgent push to move Predix forward. In this post, I’ll drill down into some specific observations.
A developer-focused conference
Running from Sunday to Wednesday, the program was frontloaded with training and certification tracks, so it made sense to arrive early to meet and talk to the people using Predix (or who at least intended to in the future).
The first noticeable thing was that the conference was unremarkable in a really important way. The general feel, the audience, the tech-focused tracks—this conference didn’t scream Giant Corporate Marketing Event, but rather fit within the norms of any open-source software conference. It even had that specific aesthetic—folks in jeans and t-shirts wearing hoodies and lanyards, sitting in windowless auditoria in soulless hotel event complexes, listening to speeches while their open laptops bathed their faces in weird blue under-lighting. Not a lot of suits. Not a lot of anything other than the session tracks themselves.
It was all about using Predix, learning Predix, and discussing plans for Predix. It was practical and hands-on, and there was a sense of urgency underlying the polite discussions of things like using image processing and time series to reduce rail accidents. The six or seven GE teams I talked with all spoke of a “push” and a “consistent drumbeat” to move everything over to the Predix platform. This was a group executing on a clear, simple directive. No one pretended it would be easy, but to a person they articulated the same basic message: Predix is the future; there is no hedge or fallback.
It also stood out how many people in attendance—GE and non-GE alike—were from outside of GE Digital, and either there to check things out on behalf of their organization or, more frequently, there to see if they should invest more resources. On the one hand, this is to be expected for attendees from a conference focused on training and certifying new developers to a platform. On the other hand, it underscores the new and unproven (and risky) nature of the Predix platform.
Predix in Action: Bright Spots And Growing Pains
A GE developer working in power generation said his thermal imaging and anomaly detection app for power plants was less than six months old. But, he quickly added, that was the point. Along with the other teams present, he was on the vanguard of an effort to port existing software (e.g. control systems for power stations, device monitoring, etc.) to Predix and to speed release cycles using the composable nature of the platform.
When asked what was missing, he was quite forthright: documents lag features, client libraries don’t work as advertised, feature roadmaps are unclear, and the API just doesn’t do everything he would need yet. These are not trivial problems—per Adam Duvander (API front-end expert), getting documentation wrong and lack of transparency are among the biggest obstacles to API adoption.
Perhaps the biggest problem the GE developer mentioned was lack of backwards compatibility: new, undocumented features broke documented ones. One can imagine the frustration a developer might experience when a previously working application no longer functioned without reason. It is the sort of frustration that leads not just to project abandonment but, worse still, bad word of mouth.
Still, these are very early days and the issues are hardly unsurmountable. I’ve never seen a project or platform where that wasn’t the case, at least at some point. The developer said two things that indicate the project has some momentum:
- The changes and improvements are happening fast. New features, fixes for breaks, the proven ability to ship something—in my experience, developers will cut an early project a little slack if the developers behind Predix show evidence they care.
- GE cares. Or, at least, GE is making a pretty big (as in company-transforming) bet. GE leadership, by requiring that every piece of software get ported and all new projects start on Predix, is using the considerable heft of their own internal capacity to force Predix to evolve and respond to user requirements by force-seeding a diverse and motivated developer community. In other words, someone at GE gets it.
This is a well-established pattern, at least in cloud. Companies productize their competitive advantage—the question is whether GE’s scale and resources are a decisive advantage or end up as a hindrance. Judging by Transform, they are aware of their challenge and trying to address it by adopting a more nimble, developer-driven approach to building Predix. Anyone wishing to create their own platform, especially one intended to be open to the public for commercial use, could do worse than emulating this pattern.
While this isn’t the post to cover use of Predix in deep detail (a post about building applications on Predix is next), it is worth concluding with a few observations about the platform itself that became evident during the week.
Security concerns and time-series analytics dominate Predix and dominated the Transform program. A quick look at Predix.io shows a handful of well-defined services already up and running for public consumption, including the recent additon of 3rd-party applications like RabbitMQ (a high-throughput message queue app) and Redis (a key-value cache). The analytics services currently available all revolve around processing time-series, mainly to detect anomalies.
Basically, every app I saw detected heat and stored time-series. Whether this is because of the limited number of available services on Predix or simply reflects the reality of industrial IoT (at least in its nascent stage) is unclear. The systems—turbines of all kinds, twenty ton locomotives, medical imaging machines—all have huge, expensive parts that move through, around, and onto each other, sometimes thousands of times a minute. Uncontrolled heat can be the biggest threat to both performance and longevity.
Adoption, thermodynamics, time-series, and building a massive-scale platform one real-world use case at a time–these were the themes of Predix Transform. In the case of the last theme, Predix Transform was about watching a plan unfold. GE is using its two key strategic assets—their size and their deep organizational understanding of managing physical systems—to do something so far that no one has really attempted at this scale. They are bridging the physical world of industrial systems and the distributed, digital world of cloud. And when every one of their systems use software running on Predix, their partners and customers will have to follow.