Traditionally everything we did involved edge computing – or the use of local computers and servers. But a number of big companies like Amazon, Microsoft and IBM convinced corporate America that there were huge benefits of cloud computing. And cloud computing spread to small businesses and homes and almost every one of us works in the cloud to some extent. These benefits are real and include such things as:
- Reduced labor costs from not having to maintain an in-house IT staff.
- Disaster recovery of data due to storing data at multiple sites
- Reduced capital expenditures on computer hardware and software
- Increased collaboration due to having a widely dispersed employee base on the same platform
- The ability to work from anywhere there is a broadband connection.
But we’ve also seen some downsides to cloud computing:
- No computer system is immune from outages and an outage in a cloud network can take an entire company out of service, not just a local branch.
- A security breach into a cloud network exposes the whole company’s data.
- Cloud networks are subject to denial of service attacks
- Loss of local control over software and systems – a conversion to cloud often means losing valuable legacy systems, and functionality from these systems is often lost.
- Not always as cheap as hoped for.
The recent move away from cloud computing comes from computing applications that need huge amounts of computing power done in real time. The most obvious examples of this is the smart car. Some of the smart cars under development run as many as 20 servers onboard the car, making them a driving datacenter. There is no hope of ever moving the brains from smart cars or drones to the cloud due to the huge amounts of data that must be passed quickly between the car’s sensors and its computers. Any external connection is bound to have too much latency to make true real-time decisions.
But smart cars are not the only edge devices that don’t make sense on a cloud network. Some other such applications include:
- Drones have the same concerns as cars. It’s hard to imagine a broadband network that can be designed to always stay in contact with a flying drone or even a sidewalk delivery drone.
- Industrial robots. Many new industrial robots need to make decisions in real-time during the manufacturing process. Robots are no longer just being used to assemble things, but are also being used to handle complex tasks like synthesizing chemicals, which requires real-time feedback.
- Virtual reality. Today’s virtual reality devices need extremely low latencies in order to deliver a coherent image and it’s expected that future generations of VR will use significantly more bandwidth and be even more reliant on real-time communications.
- Medical devices like MRIs also require low latencies in order to pass huge data files rapidly. As we built artificial intelligence into hospital monitors the speed requirement for real-time decision making will become even more critical.
- Electric grids. It turns out that it doesn’t take much of a delay to knock down an electric grid, and so local feedback is needed to make split-second decisions when problems pop up on grids.
We are all familiar with a good analogy of the impact of performing electronic tasks from a distance. Anybody my age remembers when you could pick up a telephone, have instant dialtone, and then also got a quick ring response from the phone at the other end. But as we’ve moved telephone switches farther from customers it’s no longer unusual to wait seconds to get a dialtone, and to wait even more agonizing seconds to hear the ringing starting at the other end. Such delays are annoying for a telephone call but deadly for many computing applications.
Finally, one of the drivers to move to more edge computing is the desire to cut down on the amount of bandwidth that must be transmitted. Consider a factory where thousands of devices are monitoring specific operations during the manufacturing process. The idea of sending this mountains of data to a distant location for processing seems almost absurd when local servers can handle the data at faster speeds with lower latency. But cloud computing is certainly not going to go away and is still the best network for many applications. In this factory example it would still make sense to send alarms and other non-standard data to some remote monitoring location even if the data needed to keep a machine running is done locally.
