RTEMS vs VxWorks

[gregory.menke at gsfc.nasa.gov]

Ooops, I replied directly to him and not the list.  Below is what I sent;


Carl-Johan Hansson writes:
 > Hi
 > 
 > I'm working for a company that for a long time used VxWorks as the main RTOS 
 > in their products, now I want to shift to RTEMS because it's cheaper and I 
 > don't want to give away more money to Wind River. But now I need to persuade 
 > my boss but I lack information about the differences in performance between 
 > RTEMS and VxWorks.
 > 
 > So my question is: Anyone made a compare regarding speed, code size, and 
 > memory usage?

I did a bunch of performance measurements that suggested the
performance is indistinguishable with respect to OS overhead in system
calls; the calls in question are semaphores, queues, mutexes.  This is
a measure of OS overhead that the application must operate under.

Depending on the processor, dhrystones are a bit better in RTEMS, but
Whetsones are a little slower than vxWorks.  Naturally those 2
benchmarks aren't good for predicting real-world performance, but they
do indicate a relative degree of performance.  In general, the integer
and fp performance should be pretty much the same because Wind River
uses the GNU chain.  OTOH, RTEMS stays with the latest and greatest
GCC, so any optimizations that show up there will benefit RTEMS-
whereas Wind River uses a fairly ancient gcc and binutils, IIRC.

I used to have a website with my results on it, but due to security
censorship, I had to take it down.  Joel may have it mirrored
someplace.

I don't have the facilities or time to measure the relative context
switch overhead between RTEMS & vxWorks, but its doing the same kinds
of things for both, so there shouldn't be much difference.

Memory-wise, RTEMS is pretty small, you can easily get the kernel down
to 100k or so.  Memory usage is up to you, most bsp's allow you to
specify how much memory to use, and what to use it for.

As in vxWorks, a lot of the performance you get is dependent on the
quality of the bsp, and like vxWorks, the RTEMS ones vary also.  If
you could indicate what processor and/or board you're concerned with,
we might have more specific comments.




Joel Sherrill writes:
 > Carl-Johan Hansson wrote:
 > > Hi
 > > 
 > > I'm working for a company that for a long time used VxWorks as the main 
 > > RTOS in their products, now I want to shift to RTEMS because it's 
 > > cheaper and I don't want to give away more money to Wind River. But now 
 > > I need to persuade my boss but I lack information about the differences 
 > > in performance between RTEMS and VxWorks.
 > > 
 > > So my question is: Anyone made a compare regarding speed, code size, and 
 > > memory usage?
 > 
 > I'm surprised that Greg Menke hasn't piped up by now. :)
 > 
 > http://that.gsfc.nasa.gov/osgroup/
 > 
 > has a link to their benchmarks.

So the pages are still there... they're certainly not under any
maintenance from me.  I've abandoned the osgroup webpages because I
don't have an account which is really usable for uploading new copies
of the content, all my header stuff got broken by the forcible
rehosting, and its content is subject to censorship.

As far as I'm concerned the copy you made of osgroup contents some
time ago is the standard mirror.  If you're interested in hosting it
elsewhere, I'd be glad to contribute to it as occasions arise.

On a ligher note, if I can make other measurements/benchmarks on MIPS
or PowerPC that people might want, I'd be glad to help.

Gregm