> one process per core, each locked to a single core; use locked local RAM only (effectively limiting NUMA); direct dedicated network queue (bypass kernel); direct storage I/O (bypass kernel)

I have no idea how to do any of these things. What are the system/api calls to lock a process to a kernel? How do you bypass kernel IO?