I was catching up on some podcasts during my drive to work recently and came across an interesting topic on the BBC's Digital Planet podcast. The discussion was regarding the “impact” cosmic rays can have on our desktop computers and more importantly very expensive grid platforms such as IBM’s Blue Gene. Computer crashes can be annoyng to the average user but on grid platforms running simulations that require accurate data they can be very costly. During the podcast they had interviewed with an Intel researcher (who's name I have forgotten sorry) who was discussing his strategies for counter acting the effects these highly energetic particles can have.
Before I outline his proposal it’s probably worth explaining some rudimentary physics.
What is a cosmic ray?
Answer: A cosmic ray is a particle with mass greater than zero that has been been accelerated to a speed that is approaching the speed of light.
How does it get that fast?
Answer: A range of cosmic events can cause this but let’s just say for this blog, a star exploded millions of years ago and that it is this explosion that caused the cosmic ray.
I heard that most cosmic rays just pass right through us.. How come?
Answer: Well thanks to Mr Einstein we know that with increased mass comes an exponential increase in energy in order to accelerate the matter towards light speed. Therefore cosmic rays are generally low is mass. As a result, they can pass through the atoms in your body undisturbed. (Well mostly)
So how do cosmic rays cause computer crashes?
Answer: Computers processors are based on logic gates which manage the flow of electrons on the processor. When a cosmic ray hits the processor a surge or peak of energy is releases in the form of electrons which can result in the processor providing inaccurate results. (Hence it crashes).
So with the details out of the way I can get on to explaining what the Intel researcher was planning to in order to resolve this problem. Basically, his view on cosmic crashes was that it’s going to be almost impossible to stop them from happening, insulating the processor with cosmic ray impenetrable materials is going to be very hard, very large and expensive. So, instead he proposed that what we should do is build detectors which can identify the surge of electrons associated with a cosmic ray and then instruct the processor to rerun the previous calculation that may have been affected.
I must say... I like his thinking.