At this point, I want to switch gears to more recent results. The numbers you have seen to this point were measured a while back. Since then, we have redesigned the IPC subsystem. In the course of that, we removed the fast IPC implementation temporarily, and it hasn't gone back in yet. The numbers from here on therefore reflect an IPC mechanism that is roughly 8 times as expensive as the final one will be. Bear this in mind when you look at the Linux comparisons. [The 8x statement is based on my belief that we will be able to get the new IPC to the same speed as the old one. It was certainly designed with that in mind, and none of the changes should impede the fast-path implementation.]

The numbers you will see from here on are measured on a 200Mhz PPro system. While I haven't included them, I've gotten equivalent comparative metrics on Pentium-II/233 and Pentium-II/400 machines.