Some punditry


A scribble on broad ideas of future computers

I should like to indulge myself with a brief scribble absurdly generalising some ideas I batted out with the SRCF chaps last week, in an exploratory conversation. I can’t reproduce with any sense all the thoughts I had and the subset we discussed, but perhaps this is a more sensible exposition in any case.

Motivationally, imagine what happens once we get jobs and have to pay for our own power. I would consider downgrading my computer (65W processor, low power four years ago, and equivalent parts) to something less hot. A 1U Atom computer with peak whole-system power of 20W these days can provide network storage, act as a router and so on, and play DVDs on top of that, for about £300. I can see why people get these devices.

The paradigm for a long time has been that desktops were beefy, high-power devices, and laptops weaker systems with lower power and less capability. My suggestion is that either now or in the near future we will realise that is backwards.

Consider storage: it is the only thing which absolutely limits our capabilities at the moment, given our huge demand for storage. Rather little we do has so great a ratio of computation to data that we are CPU-bound, and the disk cost far outweighs the processor cost in many classes of server. There are a lot of things which we would like to do with computers, but it storage is the only thing which actually prevents us from doing most of these things, in that the cost of the storage becomes the critical factor for many of them, and the range of projects where this is the case will only increase.

As an aside, consider where all this data is coming from. Traditionally, data has been something we create, and there is only so much of that we can manage. A drawing here or there, a letter, a document: very small demands. By contrast, the explosion in data comes when we want to store data that we have not generated, but collected by observation. The stuff we are and is around us is the richest source of data imaginable: the colour tones of faces and movement in videos, the shapes of potholes, the physical wear marks on our machinery. All around us in every direction and part of every object is data about it which we might use if it were economical to store it. So, just because video is topping out does not lessen our future demands for storage as a culture.

The old paradigm then was that the beefy systems were fixed desktops, and the laptops and mobiles were lighter. If data is the only thing that is too bulky to carry around, it will have to live somewhere, either on a storage service remotely, or in a consumer device in our houses. We cannot take it all with us though, so we will have to leave a device behind to manage it. That device will have plenty of processing power to just serve up data. In one direction then, our desktops or home servers will tend to be very low power, as the remote system performs rather little transformation on the data beyond subsetting.

In the other direction, the high power devices in our lives are the ones which process data. They can come with us, because the data is the only thing heavy enough to have to leave at home or in the data centre. They are higher power, so energy considerations would suggest that they are device which gets switched on and off a lot. This is sounding a lot like a laptop.

So, am I suggesting that netbooks and the (comparatively) lightweight model of portables is doomed? That depends. The curve that suggests storage becomes the dominant issue relative to everything else is easy to read. Extrapolating which of CPU or network growth grows fastest is a bit harder. We could end up with processing farmed out remotely too, and our portables become very thin remote clients grabbing everything we see on the screen remotely; or, mobile bandwidth for highly-responsive visual interfaces might remain a challenge for a while and the increasingly-powerful devices could generate their own pixels. Or, at least have a system where one local device powers many others so that a one mobile device can serve its interface over a short-hop network to other local devices like a car or phone system.

This is speculative and open ended. I just wish that Atom came with server-grade features like fast networking, ECC RAM, and fat drive controllers so that we can get one step closer to those sub-10W local servers I started looking at.