Cloud's Relevance To Recede As Edge Devices Take Over


The cloud as we know it is on the verge of being displaced in large part due to the coming focus on the edge as compute and storage increasingly become part of the trillion-plus devices soon to be a part of the Internet of Things.

That's the word from Peter Levine, general partner at venture capital firm Andreessen Horowitz, who told solution providers attending this week's Ingram Micro Cloud Summit X in San Diego that his views a couple of years ago on the future of the cloud elicited a strong response.

"People thought I was crazy," Levine said. "People hadn't really started moving to the cloud, and I was talking about the end of it."

[Related: Wheeling And Dealing: 5 Recent Acquisitions Disrupting The Cloud Market]

Sponsored post

However, Levine said, investors skate to where the puck is going and closely watch future expectations. "Everything gets replaced. … There are always new things waiting to take their place," he said. "And that is true with cloud."

The cloud today is really a centralized IT model where a user can go to the cloud, look for information and get a response, but this will change as IoT starts to become common, Levine said.

For instance, he said to consider a self-driving car as a data center on wheels collecting data on the edge. Such a vehicle has between 50 and 100 CPUs which, when connected together, really form a data center. Connect the cars together, and one essentially has a distributed network.

"Compute, storage, all of this will get rebuilt as this new paradigm comes upon us," he said.

Levine also said to note that, until recently, most of the information collected has been collected by humans who input the data. Now, he said, an ever-increasing amount of that information is being collected by consumer and enterprise devices, and as those devices store more data, they will need data protection capabilities.

Because of that increasing amount of data being stored on the edge, compute will have to move there as well because the decision that needs to be made may not accept the latency of today's clouds, Levine said.

"If we have to wait for a self-driving car to collect data through the cloud, good luck," he said. "It will blow through four stop signs while waiting."

Historically, the industry has moved from centralized IT during the mainframe days to distributed IT and now back to centralized IT with cloud computing, Levine said. "I'll argue we're going back to distributed," he said.

Levine said expectations are that there will be over 1 trillion devices on the edge in the not-too-distant future. "We're talking about a massive amount of devices that will have to be managed and controlled on the edge," he said.

Those devices will increasingly require machine learning to take advantage of data collected at the edge, which will change the nature of the cloud, Levine said. "The cloud becomes the last record of what happens on the edge," he said.

At the edge, devices will change in four significant ways, Levine said.

The first is that sensors will start generating massive amounts of data. As an example of that, Levine said that the self-driving car can generate 2 GB to 100 GB of data per mile. He also said to consider the amount of data sensors that become a part of people’s clothing or running shoes will generate. Those devices will be pushing data through themselves and will require massive amounts of processing, he said.

The second is that devices will start generating their own inferences from the data using machine learning, Levine said. That data is very unstructured, and will generate a lot of noise and require continuous training, he said. "Machine learning will extract data and make it relevant," he said.

The third is that devices will start to act on the data in real time because the latency between the devices and the cloud will be too high, Levine said. "So any decision will have to be made by the device," he said.

The fourth is that data sent to the cloud will be used to train the devices based on that data, Levine said. "Updates will be pushed back to the edge devices to make them work better," he said.

Levine also made three predictions of where he sees the cloud and edge devices moving going forward.

The first prediction is that the sensor data explosion will kill the cloud. "We may even retreat to, and I hate to say this, peer-to-peer computing. … I think this will fundamentally impact security," he said.

The second is a move away from today's logic-based programming to data-centric programming to support computing on the edge. Levine said that, in the future, devices will use data to solve issues, and neural networks will be built on the data. "In the future, the dated collected will be used to write the programs," he said.

The third is the massive impact coming as the price of edge devices plummet. Levine said machine learning and vision chips will become commonplace, and cited as an example how prices of LIDAR sensors used to map an autonomous vehicle's surroundings have plummeted from $75,000 to $500, and will eventually fall to below $5. "We'll see an almost unrelenting ability to place sensors everywhere," he said.

During the question-and-answer period following Levine's presentation, a solution provider asked what happens to IT without centralized services to manage.

"This paradigm shift unlocks the power of IT and the entire world becomes the domain of IT," he responded.

Levine's presentation showed yet another example of how IT is constantly changing from generation to generation, said Eric Long, president of TeraCloud, a Dallas-based cloud services provider.

"It's a metamorphosis," Long told CRN. "Whether the cloud is dead or not is subject to debate. I think it's just going to morph into whatever's next."

Levine's example of how so much data is collected from self-driving cars as to require computing at the edge to make decisions was a good one to illustrate what is happening on the edge, Long said.

"Decisions are starting to be made in the devices right now," he said. "When you consider Tesla and the decisions the car has to make ... those decisions have to be made in real time in the vehicle itself."

Levine's look at the future of the cloud makes sense from a business standpoint, said Edward Bogdan, president and CEO of Bogdan Computer Services, a Abingdon, Md.-based solution provider with an SMB focus.

"SMBs don't have the bandwidth to use the cloud," Bogdan told CRN. "Given the growing amounts of data being collected, it will be hard to push it to the cloud."