Memory Performance Issue with Fedora Core 2 Kernels

[Dave Jones]

On Fri, Jul 30, 2004 at 06:06:54PM -0400, Gene Heskett wrote:

 > >>>The problem has gone away.  I suspect that 4G/4G is broken.
 > 
 > I'm certainly leaning in that direction also.  Since turning on the 4G 
 > switch, I've had literally dozens of Oppsen

I hope you've been bugzilla'ing them.  In many cases, these are caused
by drivers doing things like dereferencing memory directly instead
of using the correct means to copy to/from kernel/user space.
Without a 4g/4g kernel, this happens to work, however in some cases,
they can point to a security hole. This code was always broken,
4g/4g just makes it more obvious.  A lot of issues have already been
caught with this that had gone unnoticed for a long time, however its
not unlikely that some remain, but without bug reports, we'll never know
and they won't get fixed.

 > >> It sounds more "broken as designed", to be honest. It sounds as
 > >> though your setup is doing a *lot* of context switches between
 > >> user mode and kernel mode. The basic trade-off that 4G/4G flushes
 > >> the TLB each context switch, so that kernel and user both get
 > >> nearly 4 GB, and TLB flushes are expensive.
 > 
 > Ouch!  Does this help explain why my old mobo runnng a 1400XP athlon 
 > could do 4.5 seti nits a day, and a 2800XP with this turned on is 
 > only doing 6?

The seti stuff should be CPU bound, and not making lots of transitions
in/out of kernel space (which is the only time you really notice
measurable overhead with 4g/4g).  If it is doing lots of syscalls,
something is wrong somewhere. Might be worth strace'ing the app and
finding out what its doing.

		Dave