As I have mentioned earlier I was having problems with
linux-2.6.14-rt6 preemption settings k3 and k4 running out of memory.
While I don't have a fix I seem to have zeroed in on the problem.
Both my embedded target boards have CF cards which are mounted
read-only. Since I am using busybox-1.01 and because busybox ships
with mkfs.minix I simply used that to create a minix filesystem on
/dev/ram0 which I mount on /tmp. My test programs create named FIFOs
on /tmp which are integral to the tests I was performing.
Normally I can't get either of the above kernels to run for more than
15 minutes without an OOM, but switching from a minix fs to ext2
linux-2.6.14-rt6 k4 has been running on one of my boards for over 2
hours.
This might not be the solution but it's looking hopeful. I wanted to
mention this now, in case anyone else was curious about this issue.
I'll let these kernels run over the weekend to see how they perform.
I'm re-running my latency tests based on 2.6.14-rt6 and will update my
report when I have those results.