Paul McKenney, IBM Linux Technology Center

Real-timee, database, and high-performance computing (HPC) developers have often asked: "Can't you get the kernel out of the way?". Recent adaptive-idle work permits just that: Linux is there when you need it, but if you follow a few simple rules, it is out (almost) of your way otherwise. This approach will provide (almost) bare-metal multicore performance and scalability to databases as well as to HPC and (most importantly) real-time applications. However, timekeeping requires that at least one CPU continue receiving scheduling-clock interrupts in the presence of any non-idle execution, which is simply in appropriate from an energy-efficiency viewpoint. Unfortunately, simple code to determine if all CPUs are idle is not scalable. This talk will give an overview of adaptive idle and outline how to scalably determine whether scheduling-clock interrupts can be shut down across the full system while avoiding any embarrassing time-skew incidents.