The shift to small, single-purpose devices (wonkish)

I was recently asked why I migrated from a NUC to a Nucleus. And I have taken positions on devices and on NAS that have been disconcerting to some. These views are rooted in a broader view of the future evolution of technology. So I’ll explain my reasoning here. It might be interesting for people who care about the tech field, and it might be useful for people thinking about setting up Roon. And if you don’t care about the technology, it might not interest you at all.

I think we are in the midst of a profound shift to ubiquitous, small, single-purpose devices. Not just a change in how computers are configured, from large general-purpose to small single-purpose. More fundamentally, small, single-purpose, intelligent and network-connected devices will be everywhere and will have a huge impact on the world, bigger than the past computer revolution. In the Roon context I favor Nucleus and MicroRendu and such devices over a Windows PC or a Mac or a NAS. But the observation applies much broader than audio. And the magnitude of the shift makes it long term: it is a prognosis for the next 50 years. But it has started.

History: the general-purpose machine revolution
Before computers, machines had a specific purpose: a drill is a drill, a crane is a crane, a refrigerator is a refrigerator. There were general purpose components, like motors, but machines were purpose-specific. The computer changed that, they were general purpose machines, even on the most fundamental mathematic level computers are general purpose.

Computing is embodied thought. You can imagine the impact this concept had on a young engineering student in the 60s: I could imagine a machine to do something, spend a few hours writing down my thoughts, and I had created a machine. And as thought is general-purpose, so are the machines we conjure.

This was a profound revolution. And over half a century, Moore’s law drove stunning improvements is price-performance, and what we learned about software engineering drove a stunning expansion in the scope of the general-purpose concept.

Management complexity
But recently, since the turn of the century, we have come to realize that systems management has become one limiting factor. And in my view, this is driven by the complexity of the general-purpose architecture.

I am particularly sensitive to this, I spent twenty years working on this problem. I saw studies by Gartner that showed the Total Cost of Ownership for a corporate PC was several thousand dollars per year, far beyond the cost of the machine. We strove to improve, but the fundamental complexity is irreducible because of the ambition to make the machines general purpose.

Roon is not corporate, management is not a cost, but the root problem remains: complexity makes it difficult to set things up and keep them running reliably and adjust and tune them. Computing complexity is the bane of our enjoyment. Just look at the forums.

In parallel Moore’s law has upended the economics. We used to think of Moore as continual speed improvement, but it isn’t a technical “law”, it is an economic observation. As it is, we can’t make machines faster like we did because they melt, but they are cheap so we can have many of them. That’s the cloud model, and it is also the new device model.

It’s the Internet of Things model, the Ambient Computing model. We will be surrounded by smart, connected devices. They will have compute capability built in, but it will be quaint to refer to “computers”.

And the changing economics will change our attitude about what resources we leverage. An anecdote that illustrates the product design consequences of price-performance shifts: in the eighties somebody invented a general-purpose kitchen appliance, by inserting an electric motor in the kitchen counter, and you would have blenders and food processors and can openers that would snap onto that motor. It failed in the market, in my view because it tried to reuse the electric motor which is no longer an expensive component. There were practical disadvantages with a single motor fixed in one location, and the economic shifts allowed us to solve that by having lots of small electric motors built into single-purpose devices like blenders and food processors and can openers.

Similarly, we will have less incentive to reuse computers for multiple purposes because they are cheap, we can have many of them. But this works only if the many are so simple that managing the swarm of single-purpose devices is in the aggregate less work that managing the single multi-purpose device.

This is profound: computers are cheap so we can have many of them. And with many cheap computers the complexity-driven management problem is a gating factor.

So this is the heart of my observation: we can have a lot of inexpensive compute power, and we will revolutionize our world, if we can solve the management problem, if we can reduce complexity, which requires single-purpose devices.

And I think this will drive a revolution that will dwarf what computers achieved so far, with vast quantities of single-purpose computers, building on but inverting the general-purpose machine revolution of the past 50 years.

Software complexity
It should be obvious that I am primarily talking about the software stack. The hardware is often general purpose, with a specific device purpose we can simplify and optimize it, but the challenges I discuss are primarily in the software.

And it isn’t that Windows or MacOS are poorly designed. They are well suited to the purpose. But that general purpose makes them complex, which makes them ill suited to the purposes we have now.

These single-purpose devices are typically based on Linux, but the point is not that Linux is easier to manage when used as a general-purpose system. Its advantage is that it can be fitted to a specific purpose, because of a modular architecture but more importantly because the open licensing model. So while Linux as a whole is complex, Roon Labs can build a simple Nucleus based on Linux.

Meridian Sooloos built single purpose devices, but the economics at the time were terrible. Roon initially leveraged the power and flexibility and economics of general purpose computing, Windows and Mac for processing, and iPad and Android for user interface.

But running a Windows or Mac machine or a NAS as a Roon server of necessity limited the customer base to the tech-savvy. It was too complicated. And for economic, not just technical reasons: a Windows or Mac machine was expensive enough that we would always want to leverage its awesome power, and use it for Office, and Photoshop, and web surfing, and backup. So we complicated the machine because of the economics.

I eventually went with a NUC. It was cheap enough I didn’t have to fall in that trap. I put Windows on it but spent a lot of effort simplifying it. And it worked well, I didn’t touch it for years, but it took a lot of skilled work to simplify it.

Then I set up a Roon server for my son, and this time I used ROCK on a NUC, and it was simpler, because ROCK was already simplified, and because it was a single purpose device, you couldn’t put Office or Photoshop on it.

But even that was demanding, most of my friends and relatives wouldn’t feel comfortable building a ROCK NUC.

So eventually I got a Nucleus. The holy grail. The simplicity of installing an FM tuner.

And where I needed endpoints, I got MicroRendu.

Both are small, single-purpose, fixed-function. Easy to install and configure. Reliable.

(There are less expensive options, more DIY, but DIY vs. turnkey is a different conversation — they are still single-purpose so they fit in this story.)

So these decisions, Nucleus and MicroRendu and such, is driven by my conviction that this is the right way. I’m a computer guy, I have spent 40+ years in the industry, I could keep general-purpose computers working for Roon, but I don’t want to. And I know most people in the world can’t or won’t either.

If we sit back and think about this shift, it is stunning. If I want to drive a sound system in another room, it would be crazy to buy a classical Windows PC and set it up only for the purpose of connecting to the DAC. But today we see Ropiee and their ilk priced like a quality cable, and they are really general-purpose computers with general-purpose operating systems but they have been simplified enough to completely change their nature. Wow.

This is technical shift, and economic shift, and most importantly a mindset shift. We are walking away from the central concept of the computer revolution.

I am convinced this revolution is under way and will have huge impact, as it spins out over decades. I am convinced enough I am making financial and career decisions based on this belief.


Maybe that’s (to become) true for the home but what’s hidden in or behind the cloud(s)? Those big compute facilities & data centers (built more and more out of “general purpose” parts thanks to “software defined” btw) we are quite literally attached to with the help of shiny small dedicated purpose gadgets with no real computing power (yet) have to be mentioned when talking about shifts. Or else a rather big part of the picture’s missing … I think. :wink:

Computing that “creates” the experience people seem to like so much isn’t really cheap. Just the end points (!) may seem to be so.

Well-said, and I couldn’t agree more. There’s a reason you’re hearing the word “appliance” a lot more in reference to these types of systems/machines. Because it’s just like buying, using any other appliance like a microwave, dishwasher, etc. Plug-n-play and purpose-built.

Yes, the technical and economic principles behind the cloud are very different.
For one thing, management is challenging but for very different reasons: the cloud operators can spread the learning over millions of systems, but scale itself is a challenge. Different attitude about failures: with millions of computers you cannot strive to eliminate failures, you have to survive failures; resilience, not perfection. I read once that in a Google data center, the meantime between disk failures is seven minutes. At home, or in a corporation, we run around and yell “omg a disk failed” and hang up a sign saying “out of order”. In the cloud, nobody reacts to this at all, automated systems mark that server as unavailable, and once a month somebody walks around with a wheelbarrow and collects the dead ones.

I was speaking only about the many billions of devices out in the world.
Those are not just dumb terminals, because a terminal is useful only if a human is there, and there are only 7.6 billion humans. The ambient computing revolution will lead to many more devices, and they will be smart and autonomous. My story about “small and single-purpose” does not mean dumb.

Yes, they cooperate with the cloud. But the contributions are on both sides.

We’ll see if “AND” turns out to be true. :wink: I’d doubt that the smartness will actually reside in the end points. It gets propagated to the end point, yes … but that’s not autonomy. Or is it?

Well, I just saw the latest Google innovation.
In photography, we got auto-exposure.
And then we got auto-focus.
Google’s new camera automates the shutter: it is looking all the time, and it contains artficial intelligence that decides when there is something interesting to shoot.
And of course, they already upload to Twitter and Facebook and Instagram.
And the social networks have algorithms that decide what to include in my feed. And probably the Like button too.
So now we humans are entirely out of the loop, the machines shoot the pictures, curate, publish, view and rate them, all by themselves.

Excellent and thought provoking post @AndersVinberg, I love reading your prose.

Anders, I’m struggling to understand what your shift from a NUC for Roon Core to Nucleus has to do with your post. Please, enlighten me. :slight_smile:


EDIT: Never mind, re-read and realised you used Windows. :wink:

Without algorithms, massive background computing going on and being always online that wouldn’t be much “fun” or wouldn’t even work. I don’t associate too much of the stuff the Letter imperium “gives” us with autonomy.

And humans are not “out of the loop” here. They are just giving away the control over their tools. A smart hammer (what ever that may be) needs a (data center) brain to drive a nail in. And it needs a human who wants (and buys) it.

I’m not entirely sure that’s good thing. Are you?

1 Like

No, it’s an example of the ludicrous use of technology.
Woody Allen in a movie a long time ago had a robocall picked up by an answering machine. He was ahead of his time.

Anyway, put aside the silliness.
There will be lots of power in the ambient computing.
In our scenarios too.

My point is that single-purpose devices makes deploying that more realistic.

It does but my reservation still stands.

Yes, my NUC used Windows.
But (a) it worked well, I had no complaints, and (b) if I wanted to simplify I could have put ROCK on it.

No, the Nucleus was for my edification, to see how far we have come.
And to inspire me in thinking about future tech.

1 Like

Just to be clear, I do appreciate the appliance concept.
Whether it’d be something SMB based or an Oracle Exadata Machine.

I still think a NAS makes a lot of sense, not only in the Roon application, as long as you use as a NAS and not for generic computing. Speaking of NAS appliance, Auspex already did that in the early 90s. :slight_smile:

The Nucleus sure looks pretty, does it perform any differently or ‘better’ than the NUC? I run ROCK on a Ethernet connected Dell PC from 2013 quite successfully. Of course, I’m starting to wonder how/if it can be improved or not.

IoT I believe is the term coined to represent all of these appliance/single purpose computers proliferating at an alarming rate. I work in technology and the security implications of non-patchable devices is daunting. Imagine all those (billions, trillions) bots to command!

Advances in computing are limited by the difficulties in implementing efficient home networks:

  • installing ethernet cables throughout your home is often not practical
  • wifi in a home often has performance problems (due to construction/architecture) and wifi mesh is costly and has limitations
  • cpl is also limited and adapters have a high failure rate (in my experience)
  • internet providers equipment is often problematic
  • home computing operating systems are not perfect (ex: windows 10 file sharing performance , Android compatibility issues, etc…)

Until home networking is improved I fear we will still be struggling to implement seamless audio solutions.

I do have a bias against using a NAS as a Roon server.

“What’s the problem, a NAS is a single purpose device, just like you say you want?” Yes, it is a Network Attached Storage device. But some people look at the hardware and say, we can repurpose this as a general-purpose server. And therein lies the problem. First, repurposing it as a general-purpose device goes against the trends I talked about, it complicates the device. Second, the people who build these are not very good at it, it isn’t a very good general-purpose server. You don’t see Merrill Lynch populating their data centers with Synology servers, they run Windows Server or Linux. So I think putting Roon on a NAS is on the wrong side of history.

It is a philosophical bias only. I haven’t tried it myself. I have a NAS but have never felt any temptation to use it for anything other than storage. I know people do it, if it works, fine. Just like Windows or Mac.

(I once worked with an experienced storage architect, and asked for a comparison of SAN and NAS. “SAN is a network built by storage guys, NAS is storage built by network guys.” She wasn’t impressed by either.)


Like I said, “as long as you use as a NAS and not for generic computing”

NAS appliances such as NetApp and HNAS makes a lot of sense as storage service such as file sharing as well as storage for app and DB servers that doesn’t need SAN speed and/or resilience. SAN makes little sense for that, both for financial and technical reasons. In fact, in goes along the lines of your philosophy since using SAN for file sharing would mean you would need to use computers with generic purpose OS as file servers otherwise.

Yes, I have shared this view, in these pages.

But I eventually got an Eero mesh, and it solved all the problems. I think the key is that there is no configuration, no settings for the user to screw up. Just like Nucleus and MicroRendu.

It is not inexpensive. But it works.

1 Like

The values of less hassle may well be worth the higher cost.

FWIW, ASUS also have mesh functionallity now. For those already invested in ASUS network devices,
it should lower to cost for Mesh WiFi quite a bit. Albeit, still a bit too much settings for users to mess up.