[rescue] sparc register windows and context switching...

[Brian Hechinger]

On Tue, Feb 05, 2002 at 09:35:30PM -0500, Greg A. Woods wrote:
> 
> It would be interesting to use something like hbench to try and measure
> actual context switching times in real OS & compiler implementations on
> several comparable systems with different CPUs and then try to use the
> SPEC or similar pure single-task CPU benchmarks to weigh the results and
> make them more useful for comparison purposes.

you come up with something that does that, and we'll add it to our little
benchmarking thing.  that would be way cool.  it wouldn't hurt to do a little
comparing between OSes as well, especially in the SMP arena, since i'd like
to see how it all pans out.  i will never forget making our RedHat/Coldfusion
x86 boxes SLOWER by adding a second CPU since the context switches went from
400/sec single-cpu to 40k+/sec dual CPU. but the same machine running FreeBSD
got between 75-90% faster with the addition of a second CPU, and the number of
context switches doubled to 800.  which is a hell of a lot better than 40K.

anyway, sorry to ramble.

-brian