The original hypothesis that the write performance
improvements cause the corruption is not correct.
Nor is it kiocluster option; just kio option can
cause corruption as well. And with Russell's mail
it appears that kiobufs may not be the source of
the problem either. So, I think we are dealing
with something more fundamental, most likely in the
I/O paths (prepare/commit write, readpage, writepage, etc.).
It would be helpful if we can find a small testcase
that shows corruption. Rightnow all we have is an rpm
build that fails; this is too complex to track corruption
directly.
If anyone has a small(er) testcase, please let us know.
--
--------------------------------------------------------------------------
Rajagopal Ananthanarayanan ("ananth")
Member Technical Staff, SGI.
--------------------------------------------------------------------------
|