> For performance, it would be good to consider shared-memory
> communication instead of pipes.  This reduces copying, which is
> important when trying to get the highest speed.

Shared memory typically requires implementing a locking mechanism, which
is not fun. The huge win with pipes is that kernels transparently do the
locking in an extremely well-standardized fashion.

Shm is good for things like blitting a display from one program to
another, in which case a bug is just a visual glitch. I think big
databases like postgres also use it, but then you have to be pretty
careful about it.

Pipes are probably good for most applications, and a shm-based protocol
should be a separate one. Perhaps shm should be done on a
per-application basis instead of generically. I don't know how to design
a good generic solution.