It seems really popular these days to say, about the future of computing, that “in a few years, you’ll have a supercomputer in your pocket."1 And it’s true: the computing power in contemporary handheld/embedded systems is truly astounding. The iPhone is a great example of this, it runs a variant “desktop operating system,” it has applications written in Objective-C, it’s a real computer (sans keyboard and a small screen). But the truth is that Andriod and Blackberries are just as technically complex. And lets not forget about how portable and powerful laptops are these days. Even netbooks, which are “underpowered,” are incredibly powerful in the grand scheme of things.
And now we have the cloud, where raw computing power is accessible and cheap: I have access to an always-on quad-core system, for something like 86 cents a day. That’s crazy cheap, and the truth is that while I get a lot for 86 cents a day, I never run up against the processor limitations. Or even gotten close. Unless you’re compiling software/graphics (gaming) the chances of running into the limits of your processor for more than a few seconds here and there, are remarkably slim. The notable exception to this rule, is that the speed of USB devices is almost always processor-bound.
All this attention on processing power, leads to predictions about “supercomputers in your pockets,” and the slow death of desktop computing as we know it. This is, while interesting and sexy to talk about, I think it misses some crucial details that are pretty important.
The thing about the “supercomputers in your pocket” is that mobile gear is almost always highly specialized and task specific hardware. Sure the iPhone can do a lot of things, and it’s a good example of a “convergence” device as it combines a number or features (web browsing/email/http client/phone/media viewer) but as soon as you stray from these basic tasks, it stops.
There are general purpose computers in very small packages, like the Nokia Internet tablets, and the Fujitsu ultra mobile PCs, but they’ve not caught on in a big way. I think this is generally because the form factor isn’t general purpose and they’ve not yet reached the commodity prices that we’ve come to expect for our general purpose computing gear.
So while I think the “how we’ll use pocket-sized” supercomputers still needs to be worked, I think the assertion that computing power will continue to rise, while the size will continue to shrink, at least for a few more years. There are physical limits to Moore’s Law, but I think we have a few more years (10?) before that becomes an issue.
The question that I’ve been asking myself for the past few days isn’t “what are we going to do with new supercomputers,” but rather, “what’s that box on your desktop going to be doing.”
I don’t think we’re going to stop having non-portable computers, and indeed, as laptops and desktops have functionally converged in the last few years: the decision between getting a laptop and a desktop is mostly about economics, and “how you work.” While I do think that a large part of people’s “personal computing” going to happen on laptops, I don’t think desktops are going to just cease to exist in a few years, to be replaced by pocket-sized supercomputers.
It’s as if we’ve forgotten about mainframe computing while we were focused on supercomputers.
The traditional divide between mainframes and supercomputer is simple, while both are immensely powerful supercomputers tend to be suited to address computationally complex problems, while mainframes are designed to address comparatively simple problems on massive data-sets. Think “supercomputers are processors” and “mainframes are input/output.”
My contention is that as, the kinds of computing that day-to-day users of technology starts to level off in terms of computational complexity (or at least is overtaken by Moore’s Law), the mainframe metaphor becomes a more useful perspective to extend into our personal computing.
This is sort of the side effect of thinking about your personal computing in terms of “infrastructure”2 While we don’t need super-powerful computers to run our Notepad applications, finding better ways to isolate and run our tasks in parallel seems to make a lot of sense. From the perspective of system stability, from the perspective of resource utilization, and from the perspective of security, parallelizing functionality offers end users a lot of benefits.
In point of fact, we’ve already started to see this in a number of contexts. First, mutli-core/multi-processor systems are the contemporary standard for processors. Basically, we can make processors run insanely fast (4 and 5 gigahertz clock speeds, and beyond) but no one is ever going to use that much, and you get bottlenecks as processes line up to be computed. So now, rather than make insanely fast processors, (even for servers and desktops) we make a bunch of damn fast processors (2 or 2.5ghz is still pretty fast) that are all accessible in one system.
This is mainframe technology, not supercomputing technology.
And then there’s virtualization, which is where we run multiple operating systems on a given piece of hardware. Rather than letting the operating system address all of the hardware at once as one big pool, we divide hardware up and run isolated operating system “buckets.” So rather than having to administer one system, that does everything with shared resources, and having the headache of making sure that the processes don’t inter-fear with each-other, we create a bunch of virtualized machines which are less powerful than the main system but only have a few dedicated features, and (for the most part) don’t affect each other.
This is mainframe technology.
Virtualization is huge on servers (and mainframes of course,) and we’re starting to see some limited use-cases take hold on the desktop (e.g. Parallels desktop, VMware desktop/fusion), but I think there’s a lot of potential and future in desktop virtualization. Imagine desktop hypervisors that allow you to isolate the functions of multiple users? That allow you to isolate stable operations (eg. fileserving, media capture, backups) from specific users' operating system instances, from more volatile processes (eg. desktop applications). Furthermore, such a desktop-hypervisor would allow users to rely on stable operating systems when appropriate and use less stable (but more feature rich) operating systems on a per-task basis. There are also nifty backup and portability related benefits to running inside of brutalized containers.
And that is, my friends, really flippin' cool.
The technology isn’t yet there. I’m thinking about putting a hypervisor and a few guest operating systems on my current desktop sometime later this year. It’s a start, and I’ll probably write a bit more about this soon, but in any case I’m enjoying this little change in metaphor and the kinds of potentials that it brings for very cool cyborg applications. I hope you find it similarly useful.
Above all, I can’t wait to see what happens.