A server that is not a PC

Today, servers are PCs. They have the same architecture as PCs. They run PC operating systems. But do they have to be PCs? Is there another approach? There might be.

First, let's consider PCs. PCs have lots of parts, from processor to memory to storage, but the one thing that makes a PC a PC is the video. PCs use memory-mapped video. They dedicate a portion of memory to video display. (Today, the dedicated memory is a "window" into the much larger memory on the video card.)

Which is a waste, as servers do not display video. (Virtual machines on servers do display video, but it is all a game of re-assigned memory. If you attach a display to a server, it does not show the virtual desktop.)

Suppose we made a server that did not dedicate this memory to video. Suppose we created a new architecture for servers, an architecture that is exactly like the servers today, but with no memory reserved for video and no video card.

Such a change creates two challenges: installing an operating system and the requirements of the operating system.

First, we need a way to install an operating system (or a hypervisor that will run guest operating systems). Today, the process is simple: attach a keyboard and display to the server, plug in a bootable USB memory stick, and install the operating system. The boot ROM and the installer program both use the keyboard and display to communicate with the user.

In our new design, they cannot use a keyboard and display. (The keyboard would be possible, but the server has no video circuitry.)

My first though was to use a terminal and attach it to a USB port. A terminal contains the circuitry for a keyboard and display; it has the video board. But no such devices exist nowadays (outside of museums and basements) and asking someone to manufacture them would be a big ask. I suppose one could use a tiny computer such as a Raspberry Pi, with a terminal emulator program. But that solution is merely a throwback to the pre-PC days.

A second idea is to change the server boot ROM. Instead of presenting messages on a video display, and accepting input from a keyboard, the server could run a small web server and accept requests from the network port. (A server is guaranteed to have a network port.)

The boot program could run a web server, just as network routers allow configuration with built-in web servers. When installing a new server, one can simply attach it to your network and then connect to it via SSH.

Which brings us to the next challenge: an operating system. (Or a hypervisor.)

Today, servers run PC operating systems. Hypervisors (such as Microsoft's Hyper-V) are nothing more than standard PC operating systems that have been tweaked to support guest operating systems. As such, they expect to find a video card.

Since our server does not have a video card, these hypervisors will not work properly (if at all). They will have to be tweaked again to run without a video card. (Which should be somewhat easy, as hypervisors do not use the video card for their normal operation.)

Guest operating systems may be standard, unmodified PC operating systems. They want to see a video card, but the hypervisor provides virtualized video cards, one for each instance of a guest operating system. The guest operating systems never see the real video card, and don't need to know that one is present -- or not.

What's the benefit? A simpler architecture. Servers don't need video cards, and in my opinion, shouldn't have them.

Which would, according to my "a PC must have memory-mapped video" rule, make servers different from PCs. Which I think we want.

The video-less server is an idea, and I suspect that it will remain an idea. Implementing it requires special hardware (a PC minus the video circuitry that it normally has) and special software (an operating system that doesn't demand a video card) and special boot ROM (one that talks over SSH). As long as PCs are dominant, our servers will simply be PCs with no display attached.

But if the market changes, and PCs lose their dominance, then perhaps one day we will see servers without video cards.

Predictions for tech in 2019

Predictions are fun! They allow us to see into the future -- or at least claim that we can see into the future. They also allow us to step away from the usual topics and talk about almost anything we want. Who could resist making predictions?

So here are my predictions for 2019:

Programming languages: The current "market" for programming languages is fractured. There is no one language that dominates. The ten most popular languages (according to Tiobe) are Java, C, Python, C++, VB.NET, C#, JavaScript, PHP, SQL, and Objective-C. The top ten are not evenly distributed; Java and C are in a "lead group" and the remaining languages are in a second group.

O'Reilly lists Python, Java, Go, C#, Kotlin, and Rust as languages to watch for 2019. Notice that this list is different from Tiobe's "most popular" -- Rust and Kotlin show on that index in positions 34 and 36, respectively. Notably absent from O"Reilly's list are C++ and Perl.

For 2019, I predict that the market will remain fragmented. Java will remain in the lead group unless Oracle, who owns Java, does something that discourages Java development. (And even then, so many systems are currently written in Java that Java will remain in use for years. Java will be the COBOL of the 2020s: used in important business systems but not liked very much by younger developers.) C will remain in the lead group. (The popularity of C is hard to explain. But whatever C has, people like.)

Fragmentation makes life difficult for managers. Which languages should their teams use? A single leader makes the decision easy. The current market, with multiple capable languages allows for debates about development languages. An established project provides the argument of sticking with the current programming language; a new project (with no existing code) makes the decision somewhat harder. (My advice: pick a popular language that gets the job done for you. Don't worry about it being the best language. Good enough is... good enough.)

Operating systems: Unlike the "market" for programming languages, the "market" for operating systems is fairly uniform. I should say "markets": we can consider the desktop/laptop segment, the server segment, and possibly a cloud segment. For the desktop, Windows is dominant, and will remain dominant in 2019. Windows 10 is capable and especially good for large organizations who want centralized administration. MacOS is used in a number of shops, especially smaller organizations and startups, and will continue to have a modest share.

For servers, Linux dominates and will continue to dominate in 2019. Windows runs some servers, and will continue to, especially in organizations who consider themselves "Microsoft shops".

The interesting future for operating systems is the cloud segment. Cloud services run on operating systems, usually Linux or Windows, but this is changing on two fronts. The first is the hypervisor, which sits below the virtual operating system in a cloud environment; the second is containers, which sit above the virtual operating system (and which contain an application).

Hypervisors are well-understood and well established. Containers are new (well, new-ish) and not as well understood, but gaining acceptance. Between the two sits the operating system, which is coming under pressure as hypervisors and containers perform tasks that were traditionally performed by operating systems.

In the long run, hypervisors, containers, and operating systems will achieve a new equilibrium, with operating systems doing less than they have in the past. The question will not be "Which operating system for my cloud application?" but instead "Which combination of hypervisor, operating system, and container for my cloud application?". And even then, there may be large shops that use a mixture of hypervisor, operating system, and container for their applications.

Virtual reality and augmented reality: Both will remain experimental. We have yet to find a "killer app" for augmented reality, something that combines real-world and supplied visuals in a compelling application.

Cloud services: Amazon.com dominates the market, and I see little to change that. Microsoft and Google will maintain (and possibly increase) their market shares. Other players (IBM, Dell, Oracle) will remain small.

The list of services available from cloud providers is impressive and daunting. Amazon is in a difficult position; its services are less consistent than Microsoft's and Google's. Both Microsoft and Google came into the market after Amazon and developed their offerings more slowly. The result has been a smaller market share but a more consistent set of services (and, I dare say, a better experience for the customer). Amazon may change some services to make things more consistent.

Phones: Little will change in 2019. Apple and Android will remain dominant. 5G will get press and slow roll-out by carriers; look for true implementation and wide coverage in later years.

Tablets: 2019 may be the "last year of the tablet" -- at least the non-laptop convertible tablet. Tablet sales have been anemic, except for iPads, and even those are declining. Apple could introduce an innovation to the iPad which increases its appeal, but I don't see that. (I think Apple will focus on phones, watches, earphones, and other consumer devices.)

I see little interest in tablets from other manufacturers, probably due to the lack of demand by customers. As Android is the only other (major) operating system for tablets, innovation for Android tablets will have to come from Google, and I see little interest from Google in tablets. (I think Google is more interested in phones, location-based services, and advertising.)

In sum, I see 2019 as a year of "more of the same", with few or no major innovations. I suspect that the market for tech will, at the end of 2019, look very much like the market for tech at the beginning of 2019.