Download the authoritative guide: Cloud Computing 2019: Using the Cloud for Competitive Advantage
Technology tandems aren’t very common but occasionally you do find two techs that can’t live without each other. The relationship between operating systems and CPUs is entirely co-dependent because they both need each other.
Another example of that is the simultaneous growth of edge computing, sometimes called fog computing by people trying to be clever, and the Internet of Things (IoT). Due to the design of IoT, it needs edge computing to maximize its potential, and both techs are in their very early stages.
Both edge computing and fog computing are strongly on the rise for the same exact reasons: an IoT data deluge. A report by Hitachi Vanata estimates that connected cars, with their continuous monitoring of all systems, will generate 25GB of data every hour.
“IoT is the next generation of endpoints,” said Lazarus Vekiarides, CTO of cloud storage provider ClearSky Data. “If you think about the sheer volume of things out there that could be generating data the number is enormous, much larger than the number of humans or cell phones.”
Mind you, at least a car has the power (via the engine) to at least handle some compute functions. A wearable device or remote sensor won’t because computation means power consumption and if it is operating on a battery, that means a shorter battery life.
So for the sake of the IoT devices, computing needs to be moved off the device and onto the server. While there are some truly massive data centers around the U.S. and a growing number around the world, they would be overwhelmed with data coming in from cars all around the country.
Edge computing, therefore, serves a second purpose – offloading the load on central data centers. Car data generated in Los Angeles can be processed on edge computing centers in Los Angeles rather than sending it to a Utah or Iowa data center. That means latency and even with the fast data centers and private backbones many are building, latency is still an issue.
“When you need to process data from millions of devices you do it in the cloud. The problem is the cloud is usually far away so that presents a fair amount of latency, and the volume of data doesn’t lend itself well to going up to the cloud,” said Vekiarides.
The design of the Internet is the inverse of IoT. The use case today is downloading something from an origin, like watching a movie on Netflix. Your home broadband connection is likely 10mbits to 20mbits down but 1-2mbits up. You click on a link on YouTube, which sends a few bits to a server, and a multigigabyte video is sent down to you.
The structure of IoT works in reverse. The endpoints are sending up massive amounts of data rather than receiving it. So the very design of the Internet is not in IoT’s favor.
Then there’s the third issue: storage. A wearable device like Fitbit is not going to store much. Even a few flash memory chips will add up to take up a lot of space and consume power. Plus, Vekiarides notes, storage has a much higher failure rate than CPUs and memory, and he would know given his business.
For something like IoT, where you collect huge volumes of data, you need to do two things: a local analysis component and central storage. So again, this is where edge computing fits the bill for IoT.
What Is Edge Computing Anyway?
Edge computing is defined by Wikipedia as “a method of optimizing cloud computing systems by performing data processing at the edge of the network, near the source of the data.” So it is about creating a sort of landing platform that’s nearby to the endpoints where all the data can be staged for some amount of computation and the latency problem can be addressed.
This means many more locations than most cloud providers have now, and for private firms it means more remote servers outside of their data center(s). Even a large scale data center provider like Equinix has only a dozen locations around the U.S. and 44 worldwide.
In many ways it looks a lot like a standard data center and uses all of the same hardware. The only real difference is its higher rate of distribution. Major data center providers like Equinix, CoreSite, Digital Reality and others are setting up dedicated hardware in their co-location centers specifically for edge computing use, and it’s the same hardware.
For business users, Amazon is about to one-up the colo providers with Greengrass, which puts an edge computing center in your remote office or location. You can deploy it on something as small as a Raspberry Pi or an x86 tower server. Amazon developed it with a number of partners, including Intel, Qualcomm, and Samsung, and uses its Lambda serverless service to deliver local machine-to-machine communication. So a remote office or factory floor can deploy their own private edge computing server for local devices.
No Standard Design
IoT and edge computing have something in common: they are both works in progress. IoT was defined fairly early but it took a little while to catch on. Early attempts, like wearables, didn’t go over well because they weren’t very good and people didn’t like wearing them. With further development, embedded devices have gotten better. Each new generation of cars become more computerized. Smartphones gain new sensors and some wearables are actually useful.
However, IoT and edge computing remain a work in progress. Vekiarides notes that it’s difficult to generalize use cases or core apps. “Every single one of these IoT apps is a snowflake, with its own unique requirements, there is no standard for IoT or edge computing yet because GM cars are unique to Ford, and factories have their own sensors,” he said.
Platforms are starting to emerge that are similar to the messaging middleware that first emerged in the 1990s, like CORBA and SOA. They are little more than standards for passing messages back and forth between devices and where the processing is done, but it’s a start.
And like every emerging standard there are lots of contenders for the crown. Gartner’s Hype Cycle for IoT Standards and Protocols lists 30 would-be standards, half of them expected to deliver some kind of business benefit. They cover IoT security, device management and sensor input.
Obviously a great deal of buildout is needed. Even with a dozen major cities covered, Equinix has a way to go, as do its competitors. For smartcars to work in major metropolitan areas, significant edge computing will be needed, whether it’s from the car makers themselves, Equinix, Amazon, or emerging micro data centers like Vapor IO and Schneider Electric.
They offer ruggedized data centers about the size of a car placed at a cell tower with a wired connection to the tower to take in 4G data sent from local devices and process it there, or forward it on to a data center. The research firm Markets and Markets believes that the micro data center sector could be worth a $32 billion over the next two years.
And more compute is needed in the data centers as well. Vekiarides said he’s heard anecdotally of some cloud customers not being able to get the compute resources they need. The problem will only worsen when millions of cars are sending data in to the data centers.
And of course, the network itself needs to get faster. Some of the major cloud providers like Google and Amazon have built their own networks for maximum throughput but for the masses, you are competing with Netflix and YouTube, which combined consume more than 40 percent of Internet bandwidth, and Netflix during peak hours is as much as 70 percent of Internet traffic.
That can be solved in part with the advent of 5G, which will be up to 10 times faster than 4G. That will take a lot of the load off the edge networks. It’s scheduled for testing starting this year, and rollout will take several years because these networks are not cheap to build.
So overall, edge computing and IoT are still in their infancy but growing together.