Hear, hear!, from the perspective of someone who's worked on an HPC resource manager, and fondly remembers data collection/analysis on real-time micro-kernel-is

Hear, hear!, from the perspective of someone who's worked on an HPC resource manager, and fondly remembers data collection/analysis on real-time micro-kernel-ish distributed systems in the '80s. I don't want proprietary software, of course; we could hack those '80s systems in interesting ways, even if it wasn't actually free software. I've sighed while removing code for systems like Irix, and Super-UX (which I thought was dead, but seems not to be). Their extended Unix resource management features were specifically useful for HPC-ish batch systems. The deficiencies of Linux are partially mitigated if you restrict the system to one job per node at the likely cost of wasted cores, but then -- like Bluegene -- you can ask what the kernel is there for. An example of the issues is lack of a sane way to manage and account memory for a job's process groups; please don't say that the cgroups mess provides that. No, OOM isn't what's wanted when you hit a cgroup limit, for instance. Then we have no real solution for security in GNU/Linux HPC systems -- or other mainstream systems (even if they did have HPC hardware support). Bluegene was interesting, but is now dead.