RTEMS vs. VxWorks

[gregory.menke at gsfc.nasa.gov]

 > > Has anyone had the opportunity to compare the task switch times and
 > > interrupt latencies of these two systems?  I have a gut feeling that
 > > RTEMS is faster, but can someone provide supporting data for this
 > > belief?

We've used a program called POSIXPERF to compare RTEMS to Lynx on a
PPC 750 running at 233mhz.  The author of the benchmark got data on
Lynx and vxWorks on a Pentium 233mhz, and using the Lynx metrics as a
baseline, we estimate RTEMS is <extremely> close to vxWorks Tornado 2
in performance.  We've not done more detailed analysis, so its
possible RTEMS is faster- but its certainly close.  

By "performance", we mean "work done in unit time".  Or, application
work done in a particular duration- this is speaking to OS
efficency/overhead in threads, queues and semaphores.  This isn't
exactly task switching or interrupt latency, but its pretty close.

http://www.ecst.csuchico.edu/~walkerg/posixperf/

I suspect variations in bsp will cause performance differences with
both RTEMS and vxWorks on a given platform.  We're getting a RTEMS bsp
for our moderately ancient MIPS satellite flight board, I know for a
fact that our vxWorks bsp has some extremely dreadful fp context
switch overhead- so its possible RTEMS will be a big benefit here- its
fp context code is far more civilized.

At some point in the next 2 weeks or so we'll publish what benchmark
data we have to date on our project pages;

http://osgroup.gsfc.nasa.gov

Posixperf is a good RTEMS demo project in several ways, so pretty soon
I'm going to upload an RTEMS-centric version of it- the author has
told me he'd be glad to have that done.  Just takes a little time....

Gregm