From the UKAuthors Collection

A 'World of Science' Special (1)

By e-griff · Originally published 17 August 2007

I am delighted to have received a special request from a Greek person who I admire very much. They asked me: ‘John, in your infinite wisdom and with your consummate ability to explain almost anything under the sun so even the dumbest bunny such as myself can understand and debate with profs ‘n boffs, will you please enlighten us all on the subject of “asynchronous processors”’.

Well, I’m delighted to do so, here ...

The World of Science – Special Edition 1

Asynchronous Processors

Of course the ‘processor’ we are talking about is the black ‘chip’ in the heart of any computer, that does all the putting together and deciding that makes the computer work. It uses electronic logic to do hard sums, but also controls what is going on, pulling in the spell-check, say, and telling the graphics card to get itself in gear. It does this in a series on ‘bits’ of information – billions of them.

All current processors are ‘synchronous’. Let us imagine it as a vast hall full of clerks at desks, with other clerks running between them. Bits of paper are being passed round all over the place and at the end of the hall is a big clock, which, at strictly regular interval, ‘bongs’. The clerks can only do something when the bong happens, and can only do one action per bong, eg the runners can only run when the clock bongs.

Now, some clerks get more work then others, so at some desks, the ‘in’ trays are full, at others they are empty. For the former, the clerk can only take one piece of paper at each bong, can only add or alter a word at the next, can only pass it on to the ‘out’ tray at the next. In the latter case, the clerk leans forward, but there is no paper in his tray, at the second bong his pencil comes up and hovers over a blank desk, at the third, his hand moves from the blank desk towards his out tray, but there is no paper. Still, he moves according to his training, working just as hard as if there were paper there.

The only way to increase the output of the clerks with this synchronous model is to speed up the clock. This means the overloaded clerks work faster and the bottlenecks clear quicker, but it also means a whole load of unoccupied clerks are also going through the motions faster as well. Let’s say their employer pays them for every bong, whether there is paper there or not. To improve the state of the overloaded clerks costs him loads, as he has to pay the underloaded majority as well, who don’t add to the output.

Now, the ‘asynchronous processor’: Imagine we scrap the clock and devise a way for the overloaded clerks to go as fast as they like when they need to and not wait idly for each bong. We also allow the underused clerks to sit still when no paper arrives. We only pay for the movement of paper, not the regular actions of all the clerks. So the overloaded clerks work like buggery and are paid for it, while the others just sit and don’t get paid.

In computer terms, the rate of paper output is the processing rate, the wage bill is the energy requirement to run it (and the heat that’s needed to be removed so it doesn’t overcook). Asynchronous processors therefore would give much enhanced output rates (no waiting for the clock at the bottlenecks) and lower energy/cooling requirements on the chip. Hooray!!!!!!

But of course, you might think my trademark scepticism has fallen victim to the thralls of scientific journalism – that I believe the vaunting ambitions of the ‘asynchronous’ crew. Never! It’s just another scientific chimæra. In practice, as we all know, the underloaded clerks will complain bitterly, sitting twiddling their thumbs while their overloaded colleagues earn pots of money. What they’ll do is one of two things (or both) – first, sabotage the work of the others, tripping runners attempting to deliver papers, or hiding the paper that they themselves get. Second, they will pass papers to each other, changing and amending back and forth, until a whole industry of recycled, ‘false’ work is being done and wage bills rocket while efficiency plummets.

The powers-that-be may then decide to offer basic regular payments to them, so the underloaded ones get at least a good basic wage. Then the overloaded ones will feel they are getting the rough end of the deal, working like buggery while the others sit back, maybe opening their own private processing businesses on the side, and getting paid for doing nothing. They will slow down deliberately. So in the end, based on my model, the asynchronous chip will end up slower and more energy demanding than its predecessor. Can this be true? All I’m saying, you scientists, is be very careful.

if you liked this ‘special edition’ please say so vociferously, and feel free to ask for your own questions to be answered. Part 4 of ‘The World of Science’ will appear soon.

Archived reader comments

These comments are preserved as part of the historical archive and are read-only.

artisus

A World of Science Special (1)

I have to think now, but this was a very good thought-provoking read!

royrodel

A World of Science Special (1)

hmmmmm

the authors a bitch or is he?

input , output

production of what???? answers?

In India they just use the Dalits

But riddle me this would it be quicker to download 10 thousand films from London to Edinburgh or transport them by a white van man.

RODEL

Around the UKAuthors table

Member Shoutbox

Say hello, share a thought, or tell fellow members what you’re working on.

Active in the last 30 minutes Checking…

Loading the conversation…