Linux hardware

LINUX HARDWARE

To bring our series to a close, Mike Bedford looks at what might be Linux’s greatest success stories: the behemoths of the computing world.

OUR EXPERT

Mike Bedford hasn’t had the chance to use a supercomputer. If anything, though, since familiarity supposedly breeds contempt, this has done nothing to diminish the wow factor he feels when delving into the world’s latest and greatest.

This article brings our series of articles on The Hardware That Made Linux Great to a close, but it’s going out with a bang. Our subject this month encompasses several processor families, including the x86 that gave birth to Linux, but the common denominator is their application in high-performance computing. In referring to HPC, as it tends to be called among those in the know, we’re talking about supercomputers, so we’re going to be looking at the absolute top echelon of computing. In other words, we’re going to be delving into computers that are upwards of a million times faster than the PC sitting on your desk. What’s more, this is one class of computer where Linux can be considered as nothing less than essential. And our justification for saying that? Well, each and every one of the world’s top 500 computers boasts our favourite operating system. In fact, none of the 500 fastest computers have used any other operating system for the last seven years.

Introducing supercomputers

In the world of supercomputing, a machine’s place at the top table is assured by its inclusion in the Top 500 list. Published in June and November each year, the list at http://top500.org shows the world’s fastest computers in order of their performance. Specifically, the ranking is dependent on performance as measured by floating point operations per second – FLOPS or flop/s. As this is being written, the current list, which appeared last November, is the 64th, the first having been published in June 1993. As you can no doubt imagine, those 32 years have shown massive performance gains. The first ever Top 500 list was headed by a CM-5 supercomputer made by Thinking Machines Corporation and installed at the Los Alamos National Laboratory in America. It used 1,024 processors and clocked up 59.7 gigaFLOPS. We’ll take a better look at it soon, but the machine at the number one position in the most recent list boasts a maximum speed of 1.742 exaFLOPS. Computer users are familiar with large numbers, but if you’re hazy about the really large ones, an exaFLOPS is a thousand petaFLOPS, a petaFLOPS is a thousand teraFLOPS, and a teraFLOPS is a thousand gigaFLOPS. The upshot of this is that the world’s current fastest computer is almost 30 million times faster than 1993’s greatest. We could describe that growth as Moore’s Law-ish, in that the true law referred to the number of transistors on a chip. Like Moore’s Law, though, it’s exponential, which means that numbers double over each equal time interval. The doubling time for supercomputer performance has been about 14 months over the 32 years of the Top 500 list – interestingly, that’s faster growth than we have seen in the performance of processor chips. Here, the period appears to have been, very approximately, every two years over the same 32 years, fuelled initially by increasing clock speeds and more recently by increasing the numbers of cores.