Google brings machine learning data processing to IoT edge environments

Google is bringing its internet of things (IoT) data processing capabilities to edge environments to accommodate enterprises that are unable to rely on cloud for latency or regulatory purposes.

On the second day of the Google Cloud Next 2018 conference in San Francisco, Google revealed how it intends to bring machine learning and artificial intelligence (AI) data processing to edge environments using custom-built chips, new hardware and an expanded software stack and service provision.

As such, the search giant debuted its purpose-built Edge TPU ASIC chip, which is capable of running machine learning models, based on Google’s Tensorflow technology, in edge environments.

During the second day keynote, Injong Rhee, vice-president of IoT at Google Cloud, said the chip has been optimised to provide the maximum performance per watt, and to help make the cost of managing IoT deployments more manageable for enterprise users.  

“We see there are emerging needs for edge computing, which is essentially running data analytics and intelligence services in locations where data is collected or what we call at the edge,” he said. “This is important as sometimes moving all data to the cloud from sensors can be very expensive.”

The setup is also underpinned by the Cloud IoT Edge software stack. This can run on Android Things or Linux-based devices and equips them with the capabilities they need to carry out machine learning-related data processing tasks.

Rhee said the setup has already been adopted by the IT services arm of electronics giant LG, which is using it to cut the amount of manpower needed in its product testing procedures and predict it could help the organisation save the organisation around $1m a year per product line.

The combined offering is an expansion of the Google Cloud IoT Core managed service, which allows enterprises to collect and process data from millions of internet-connected devices worldwide, and entered general availability in February 2018.

Edge IoT Core applies similar principles to environments and scenarios where internet connectivity might be patchy, making it difficult for enterprises to access the cloud-based data processing resources they require or where regulatory compliance precludes from using them.

“It really opens up a lot of applications to run that intelligence closer to where it is needed the most,” said Antony Passemard, head of product management for cloud IoT at Google, during a pre-keynote briefing with the press.

“Customers were telling us that [when the] internet is down they still need to apply intelligence there or [they] can’t wait for a 25 milliseconds round trip, [or they] need to have something in 20 milliseconds. So that happens at the edge.”

According to Passemard, the release marks Google out from its competitors in the public cloud, as it is the only provider to offer a fully integrated hardware and software stack to tackle data processing at the edge that is able to tap into the latest thinking in machine learning and AI.

“We are the only cloud supplier today to have combined our AI research, our software research and our hardware research to build together the hardware that is optimised for the latest and greatest AI technology in the cloud,” he said.

“This is, we think, a game changer for IoT. This is really bringing intelligence to the edge in an efficient manner,” he said.

That said, Amazon has offered similar capability – in the form of its Snowball Edge appliance – for some time now, and Microsoft has a play here, with its Azure IoT Edge offering.

In that regard, Google could be considered late to the edge computing party, said Nicholas McQuire, vice-president of enterprise research at IT market watcher CCS Insight, but the market is still in its infancy so there is plenty of time to catch up.

“Google Cloud has been a bit late to the party in the internet of things and edge computing, but [these] announcements indicate that Google is now taking this market very seriously,” he said.

“While it is to an extent playing catch up to AWS and Microsoft, the market is still in its infancy and the fact that key elements from across Google are coming together to tackle the opportunity – Android Things, chip engineering, security and Tensorflow – it is an ominous sign for the competition given what Google can bring to the table in helping firms run machine learning at the edge.”