Building a Supercomputer of Things?

Alasdair Allan
2 min readNov 9, 2017

Despite the prevalent view, not every bit of computing is moving into the cloud. In the last few months we’ve seen the start of a sea change about how we think about machine learning, and how the Internet of Things might be built, with the potential to put the smarts on the smart device, rather than in the cloud.

While building a Beowulf cluster from off the shelf hardware almost became a right of passage for computing researchers in the late nineties, the relatively recent arrival of cheap commodity single board computers like the Raspberry Pi has meant that building clusters isn’t a dying art any more.

So it was only a matter of time before these two things came together to try and build a distributed cluster of smart devices.

The #Beowulf.1 Computes Cluster. (📷:

The Computes platform is designed to run on on any computing platform, from your web browser, to microcontroller boards, to smart meters and other industrial Internet of Things devices, through to cloud installations using AWS Lamda or Azure.

“Most computers, mobile devices, and IoT devices are idle 80% of the time. We’ve developed an inter-connected, massively-parallel, globally (or edge) connected supercomputer capable of distributing computations and data across all cores connected to the Computes OS mesh network. ” — Chris Matthieu

Individually most Internet of Things devices have a pitifully small amount of computing available, and they may suffer from poor and intermittent network connectivity. However embedded devices are ubiquitous enough that a new generation of malware is now targeting them, and forcing them to mine cryptocurrency.

For those of us old enough to remember SETI@Home and other similar screen-saver based distributed computing platforms that were briefly popular ten years or so ago the idea of using “spare” computing resources will be rather familiar.

“Each node on the mesh network is essentially a kernel and a core meaning that it can request computations from other nodes or perform computations requested by other nodes. Each node includes machine leaning tools such as brain.js and synaptic.” — Chris Matthieu

Demo of and Johnny-Five to toggle an Arduino LED on and off. (📹: Chris Matthieu)

But perhaps the most striking thing about the ideas behind Computes is the parallels between the mesh-network based Computes platform, and the peer-to-peer distributed ledger ideas currently championed by the blockchain and cryptocurrency enthusiasts.

You have to wonder, perhaps, whether a distributed ledger should after all be based on a decentralised and distributed computing platform?