Why is Mach on the PA so slow?
Welcome to Mach! A large component of our Flux Project is involved with making a micro-kernel fast without sacrificing the modularity and protection aspects. However, our system is slow even by Mach standards due to lack of tuning in both the kernel and Lites and for one major technical reason. That reason is that the kernel is optimized for migrating RPC and the old, hand-crafted, special-case, send+receive fast IPC path is gone. Since even when the new RPC is configured in, some important calls do not yet use migrating RPC, they wind up on the slow, general IPC path.
