Compatibility of Solaris 8 vs. Solaris 10.

Question

We are getting core dump randomly in Solaris 10 box. The bin(s) & lib(s) are generated (compiled) in Solaris 8 box.
a) Are the Solaris 8 bin&lib(s) are compatible in Solaris 10.
b) If not in direct Solaris 10 box, but atleast can we try with container in Solaris 10 box.
Can you please let me know where the SunMicro stands on these compatibilities?
I guess the jar/class files shouldn’t be problem but mainly it’s about C/C++ files.
Please let me know if the core and/or box architecture details will help. :-)
AshokMohanty

Answer 1

A mismatch between the compile-time directives on the build machine and the execution environment seems like a promising place to look. The elf_xx calls in particular suggest your code and your platform don't see eye-to-eye on the section mapping of the process to virtual memory.
Of course, it's also a cool place to look, so there may be a bit of tunnel-vision in that estimation.
If root causes interest you, this seems like an outstanding opportunity to use DTrace and speculative tracing. Using the fbt provider, you could trace the elf_* calls of your application. You could discard any records upon exit from the given function -- those which work fine -- but commit to recording buffer the calls that don't finish. So long as you don't commit to the tracing buffer the data you can't use, the probe effect (i.e. performance overhead) should remain small.
The OpenSolaris DTrace wiki has a good chapter on speculative tracing. There are also three example scripts in /usr/demo/dtrace that use speculative tracing: errorpath.d, specopen.d, whoqueue.d. These should give you a good sense of how to proceed if you want to nail down root causes.
Regards,
Michael