Internal Linux Distro

On the way to building the large internal AMD we used engineers desktops as additional compute capacity. This was a time when AMD was perpetually behind Intel, and relatively underfunded. We needed ever single advantage we could get.

At the time (circa 2001) we were using Red Hat 6, with Linux 2.2 and glibc 2.1. We needed to deploy a consistent Linux kernel and environment on engineers desktop, to provide reliable compute jobs overnight or when the workstation was idle. And we found that consumer-grade PC hardware and especially graphics cards changed at a rapid pace. More rapidly than Red Hat updated the drivers.

My solution was to create an internal Linux distribution. This is when I learned to really respect the efforts and hard work of others who managed such complex release management efforts. We used this internal distribution, built on Red Hat 6, up until the AMD Optern was released and we migrated to 64 bit Linux, the 2.4 kernel, and the new glibc.

I was invited to give a talk to the high performance computing team at Los Alamos National Labs - which was flattering, of course. These folks were simulating nuclear weapons! They also showed me how they updated the kernel to detect multi-bit memory errors. I returned to work in Austin, and the very smart developer Donald Zoch soon implemented this same functionality for our internal distro.