Linux supercomputers

Mike Bedford explores the world’s fastest supercomputers to find out how Linux came to power every top 500 high-performance computer.

T wice each year, in June and November, the world’s fastest 500 supercomputers are revealed. Eagerly anticipated by the high-performance computing (HPC) community, the Top500.org list paints a picture of computing at the top end, and the difference from run-of-themill PCs is stark.

Perusing lists from the list’s 28 years highlights various trends, including one that will be of particular interest to Linux Format readers. From the dominance of Unix in the early days, a remarkable change has taken place more recently. After making its first appearance in 1998, Linux was powering the majority of the top 500 computers by 2004, it became universal in 2017, and it has remained the only operating system used in the planet’s top supercomputers ever since.

Why is Linux the dominant operating system in high-performance computing? That’s a question we aim to answer here as we take a look at the history of Linux in supercomputing. We can’t talk about the upper echelons of computing without also talking about the hardware, though, so be prepared for your jaw to drop as we also highlight some facts and figures about the world’s fastest calculating machines.

Linux’s super debut

In 1998, almost a quarter of computers in the Top500 list were manufactured by supercomputer specialist Cray Research, and the remainder of the list was dominated by big-name supercomputer manufacturers such as HP, IBM, NEC and Fujitsu. In other respects, though, highperformance computing in the late 1990s exhibited some considerable diversity.

Back in the 90s we saw machines empowered by a wide range of processor families including Sun SPARC, MIPS and IBM POWER in similar numbers, with DEC Alpha and HP PA-RISC also being represented. In marked contrast, over 90 per cent of all machines in the most recent list use x86 processors that aren’t too different from that those power run-ofthe-mill PCs. Even in the 1998 list, though, several supercomputers used chips that were also found in desktop PCs, albeit topend machines, and this brings us to the Avalon Cluster, which made its debut to the Top500 list in June of that year.

That machine was very different to most of the planet’s fastest computers. It wasn’t built by a specialised supercomputer manufacturer, but by a team at the US government’s Los Alamos National Laboratory. Not only that, it was also put together using off-the-shelf DEC Alpha-based desktop PCs as its fundamental building block.