Jun Sun wrote:

> On the preemption latency I am looking for the long execution pathes in
> syscalls for some typical benchmark workload.  (netbsd kernel is not
> preemptible, right?)

Right.

> Since the workload works on both Linux and netbsd, it would be
> interesting to do a side-by-side comparison between them.  Mvista.com
> used to have some numbers posted on their website for Linux, but I can't
> find anymore.  But it is easy in general to find out those tests and
> numbers for Linux.

Another company which claims beter results then MontaVista is TimeSys 
(see there side for more info - www.timesys.com).

If your targets are x86 processors also Solaris may be worth looking, 
because the Solaris kernel is preemptive and has other real-time facilities.

--- willy