Has anyone actually managed to get lmbench to work on qnx4? Just wondering
if I’m attempting the impossible.
Thanks,
Karen
i’ve run parts of it on both qnx4 and of course qnx rtp.
there are a lot of non-posix things in the code… plus a lot of what they test is
more desktop than embedded environments, and most is targetted towards monolithic
kernels.
for example, they mainly test throughput via:
pipes, messages queues, socket services
which are okay ipc mechanisms but in a microkernel os where we externalize those
services outside of the kernel, we will be slower than a monolithic system.
they also don’t use mutex’s. instead they use named semaphores which are also
(the named variety that is) sent as messages to the mqueue manager.
personally i don’t like the tests. they make assumptions that i don’t like.
for example, i can do an uncontested thread-thread mutex lock and unlock on a
ppc7400 @ 200Mhz in around 190 nanoseconds. do they measure that? no.
i prefer realworld benchmarks where you can use as much as possible the services
that are optomized for the os being used. in qnx that is mutex’s in rtp, or
unnamed semaphores. as well as native message passing or read/write api calls
(which resolve to message passing). and where speed is key, using shared memory
with semaphores/mutexes to handle contention.
Karen <chibiusa@mtu.net> wrote:
Has anyone actually managed to get lmbench to work on qnx4? Just wondering
if I’m attempting the impossible.Thanks,
Karen
–
Randy Martin randy@qnx.com
Manager of FAE Group, North America
QNX Software Systems www.qnx.com
175 Terence Matthews Crescent, Kanata, Ontario, Canada K2M 1W8
Tel: 613-591-0931 Fax: 613-591-3579
Previously, Karen wrote in qdn.public.qnx4:
Has anyone actually managed to get lmbench to work on qnx4? Just wondering
if I’m attempting the impossible.Thanks,
Karen
has source that compiles in QNX4.
Andrew