On Wed, 3 Jan 2001, Justin Mason wrote:
> Just to pick you up on a minor point -- in QNX at least, message passing
> *is* performed by the kernel.
(Which won't work on the long run, of course, as you wouldn't want to
wrap us overhead about what hardware does in ns).
The leaner the kernel (why, QNX or L4 could comfortably live in L1 cache),
and less context to shuffle in case of a context switch the shorter the
response latency.
> This is probably just a terminology point though, maybe other microkernel
> architectures call it something else ;)
I think under nanokernels you'd also want to do message passing via
userland code, especially if you're using the higher performance
interconnects (Myrinet & Co.).
This archive was generated by hypermail 2b29 : Fri Apr 27 2001 - 23:17:54 PDT