MADDOG'S DOGHOUSE

The prohibitive expense of early proprietary supercomputers is just part of why Linux has taken (and kept) the lead in this area.

- BY JON “MADDOG” HALL

A natural progression

There's been a lot of discussion recently in social media about high-performance computing (HPC) and the viability of Linux (instead of other operating systems, such as Microsoft Windows and FreeBSD) for HPC. I thought I would use this month's column to explain why all 500 of the fastest computers in the world run GNU/Linux.

Around 1995, the supercomputer market was having some issues. Specially designed supercomputers were very expensive, mostly because companies (e.g., Cray, HP Enterprise El Capitan, CDC 6600) would each spend millions of dollars on specialized hardware design, their own proprietary OS, and perhaps their own compilers (etc., etc.) to make their own supercomputers — and then sell a relatively small number of them. These companies were going out of business.

Then two people at NASA, Dr. Thomas Sterling and Donald Becker took the concept of decomposing large problems into many smaller ones and using commodity hardware (desktop PCs and Ethernet networking) to make something they called Beowulf systems.

Even buying new PCs allowed you to get supercomputing performance for about 1/40th of the price of a specially built supercomputer. Or, looking at it another way, you could get 40 times the performance for the same money.