The AWS Outage and the Need for “Offline” IoT
Thoughts on mobile offline machine learning in the wake of a “cloud” debacle.
Apparently, when the Internet breaks, your smart home can break with it. That’s what we learned from the widespread Amazon Web Services (AWS) outage last week. Last Tuesday, Amazon’s cloud services platform failed, affecting hundreds of thousands of companies that depend on it. It’s bad enough when websites go out of service, but the problem is much greater when smart home and Internet of Things (IoT) devices are involved.
We need to create better ways for smart devices to still work intelligently while offline.
Our Reliance on “the Cloud”
Most of the apps I’m aware of use AWS as their main solution for servers, hosting, databases, and more. This makes sense because S3, EC2, and the rest of AWS offering are top-notch. But relying too much on one system makes these apps and devices vulnerable. Imagine a garage door, a thermostat, or a health/medical device that relies solely on the cloud. It’s not ideal.
Today, the way connected devices work is essentially a bunch of ones and zeros run between the device, its app, and the cloud and tell the device when and how to operate. If the cloud fails, everything else fails with it. We don’t just want an IoT device to keep working when the cloud fails, we want it to work intelligently.
We don’t just want an IoT device to keep working when the cloud fails, we want it to keep working “intelligently”.
For example, if there’s a cloud outage, I don’t want my smart thermostat to just stay operational. I want it to still automatically keep my house at a cool temperature while I’m sleeping and a warm temperature as I’m getting up in the morning.
Moving to Mobile Offline IoT
As mainly a backend developer, the AWS outage made me think about my total reliance on cloud services. It also made me think about the Android and iOS developers I work with at Neura, who have to think about being offline as a valid and frequent status. At Neura, we’ve designed part of our core machine learning logic to run on the mobile device so it doesn’t totally rely on a server connection.
At Neura, we’ve designed part of our core machine learning logic to run on the mobile device so it doesn’t totally rely on a server connection.
Thankfully, we weren’t affected by the AWS outage. But even if we had been, we still would’ve been able to keep some of our operation running. AFP recently featured Neura in an article on this very topic entitled “Offline AI Revolution Awaits Smartphones.”
As smartphones keep improving in processing and computational power, it’s getting easier to put more of Neura’s artificial intelligence (AI) and machine learning abilities into users’ hands. Even though I’m a backend developer who lives and breathes servers, I’m still a big proponent of creating event detection and notification loops that run fully offline.
We increasingly rely on cool technologies that connect our homes, cars, medical devices, etc. to the Internet. When the cloud fails, we need smart home and IoT devices to still operate at their full capacity. With the increasing processing power of mobile phones, and the AWS outage as a catalyst, I expect we’ll start seeing more innovation in mobile offline machine learning.