So, this small blog deals with computer performance. I work with computers for 20 years now, trying to squeeze more effective performance from them. It all started in earnest with my masters thesis in 1996. My extension to a parallelizing Fortran compiler and runtime library (similar to what became OpenMP) was aimed at hiding pagefault costs on a shared-virtual-memory system running on the Intel Paragon at the Research Centre Jülich.
The Intel Paragon was a really great machine in many (not all) aspects. It’s most prominent feature were the LED panels within the cabinet front doors: for every one of the 48 CPUs within a cabinet, and the 2D-Mesh bidirectional communication network, LEDs would show in realtime the load of the CPU (20% steps) and the message flow. Incredibly cool to watch at – you basically could see whether your parallel code was effective or not!
For that reason, I chose a part of a picture of the Intel Paragon XPS/35 running at ORNL as the header image for this blog.