Thanks for the reply! As someone involved in computing research (sometimes OS kernel research, even) I can appreciate the need to actually get something going.

Thanks for the reply! As someone involved in computing research (sometimes OS kernel research, even) I can appreciate the need to actually get something going. Ideally, if I was starting off a project with the intent of focusing on something like distributed data management, I'd want to sit down and figure out what I actually need and write a kernel to do that, rather than pulling in Linux with its DECnet drivers and 300+ system calls and what-not. Maybe a Unix-like system with POSIX calls isn't the best way to approach the problem. The problem is that you end up spending a lot of time dicking around with the scheduler and figuring out why memory allocation sometimes screws up, and less time implementing the real deal. I've had good experiences working with a 64-bit Plan 9 kernel, since I was already familiar with the code and it was quite minimal, but I still ended up fighting the compatibility/driver crap you mentioned. It's also pretty easy to get functionality, but getting functionality and performance is a pain.