>Note that that diff does the converse of what you requested, pegging a >thread to a CPU instead of banning the CPU from running other >processes' threads.
True, but this is good starting point. >On your bigger question: I don't know of any one working on making >OpenBSD a realtime OS in the sense of providing latency guarantees. >This would require massive changes to all levels of the kernel, from >interrupt handling/routing/blocking to the buffer cache and UVM >subsystems to filesystems. Real time is a plus, but I understand that this may be massive task to accomplish and when looking for safety perspective, isolation and correctness are more important. Real time requirements come when correct working has requirements to latency. If isolation can be done and process running isolated on own core, it is good point to start to study how to make it work on real time. It is also ok that real time works partially and fails on certain I/O. As some functions are "thread safe", it is work to making some functions "latency safe" or "verified correct". Now there is OpenBSD core witch is very clean and secure, and ports where rest of packages are dumped. This kind of work requires more granularity as QA perspective.
