As our team presented the future of battlefield computing to a group of war fighters at 7:00 AM, they had to take a break because the mother of all centralized computing paradigms went down.

Amazon crashed a data center and everything from banks in Europe to Disneyworld went down – many people learned painfully the service to which they subscribed was tied to AWS – and they went down with the ship.

Fractal received an interesting phone call later in the day.

One of our larger customers – for whom we run parallel Fractal systems to their production systems – called and said their Fractal systems never missed a beat.

They stayed up even though they keep their customer data for their 4 largest, most critical customer-facing legacy systems on AWS.

Fortunately they are halfway through a Fractal implementation and have Fractal systems running in parallel with their AWS systems – and not a transaction was lost – not a customer was miffed.

Look, crashing data centers have been a phenom for decades – only made more painful because cloud companies magnify the pain by moving many companies and agencies to one large data center cluster.

The marketplace – led by government battlefield applications – is dealing with the question – centralized compute or distributed compute?

Today, for the first time in a generation, distributed technology is taking the forefront – at least in the planning of the most critical war fighting systems.

After today, corporations will learn – again – the dangers of all those eggs in one basket.

Data centers will always crash – there are more to come, certainly.

A truly distributed, quantum speed system, with NO CENTRAL POINT OF CONTROL – is needed now, more than ever, because of the real time requirements of daily life.

Sensors, IoT devices, warfare machinery like drones and satellites – cannot coexist painlessly with a central command and control structure.

Sensors, meters, control devices appear on millions of surfaces – as the required feedback loop for everything from electric vehicles to flying drones – and they are always active – continually churning out information – and they cannot have a central point of failure.

Today, a significant fraction of the planet was able to experience the inevitable result of single points of failure.

Let’s leave AWS alone and consider the distributed digital future – here is what it looks like – and it offers advantages far beyond 24/7 up time.

Truly distributed computing means the application has no central point of failure.

To make that happen, each node in an application must neither share resource nor be dependent upon any other node.

To build such a distributed system may be possible today – but it cannot scale.

We know a bit about that because governments are building the ultimate in distributed systems – drone swarms.

Making 11 drones work as if they were a small bird flock can be done – but it cannot be done at scale – as in hundreds of thousands to millions of drones. Again, we know that because the people trying to do this – with billion dollar budgets – tell us it is so.

They tell us about their attempts as we educate them on how Fractal – a fully distributed architecture – operating at quantum speed – on current hardware – can do the 100,000 drone thing – today.

The “secret sauce” is system design.

There are no specialized, optimized boards, no photon level quantum computers – just very intelligent system design. And a lot of trial and error that it took to perfect that system architecture.

A fundamental principal of true, independent distributed compute is smallness.

Every application is a tiny fragment – we call them Fractals – a tiny fragment of the DATA for the entire application.

It may be a tiny data fragment of the whole – but it is the entire APPLICATION. All of it.

Think that through.

A massive billing application – taking 90 hours to run on an Oracle Cluster – is made to run on a hardware box you can hold in your hand – in 5 minutes – eliminating the need for a central data center and all of its electric energy.

Each small application instance has the core logic of the billing system. The code runs in many Fractal instances – which aggregate to deliver a response – a bill.

Each fragment, each Fractal, handles a small part of the data – perhaps 20,000 records out of hundreds of millions.

Each tiny application instance has been optimized for what the engineers call “locality of reference.” To those not engineers, it means data is intelligently placed exactly where it is processed.

How come?

What do you think a data center, or a computer which is a subset of a data center – does all day?

It does NOT compute.

Compute is a tiny part of the data center load – the vast majority of the load is moving data to the CPU core where it will be processed.

We call that I/O or input/output.

Tiny applications, married to data that is stored where it need to be processed and in the form in which it needs to be processed, become I/O optimized – they do very little I/O and can do mostly compute.

These applications thus run from 1,000 to perhaps a million times faster than traditional applications because they are actually doing mostly compute instead of mostly doing I/O.

If you optimize I/O – disruptive things happen.

That is the Fractal equation – if an application runs 1,000 times faster, you need only 1/1,000th the hardware and 1/1,000th the power – thus no additional data center for that A.I. project.

Remember where we started.

We are talking about why distributed computing is replacing, and will continue to inexorably replace, centralized computing.

Some of the reason is that data centers crash – and distributed systems do not – because there is no there – there.

There is no centralized failure point.

Because these Fractal systems run so fast, can be built so inexpensively, they need neither the data center nor its energy – they can be run in triplicate – which all of them do.

So when AWS goes down, Fractal does not miss a beat – as happened today – and we will continue to educate customers about how to move their production systems to distributed computing – so they can wake up on a Monday and get fun calls, not panic calls.