--On Monday, February 03, 2003 13:47:19 -0800 Andrew Morton
[off-list ref] wrote:

Sorry David, I just haven't had time to play with this.  I did some quick
testing on uniprocessor shell-script-intensive loads and saw no
bottom-line change at all.

What load did you test with?

I used a simple test program that forks a null child and waits for it to
exit.  It does it multiple times (default 100) and times the aggregate
time, then computes an average.

Like I said, I saw roughly 10% improvement in that test with my patch.

I'm surprised that shell scripts wouldn't show an improvement.  I expected
they'd be more sensitive to exit performance, given how they highlighted
the performance issues with shared page tables.

My original reason for attacking clear_all_pages was because it kept
showing up in my profiling as significant, and a quick examination of it
looked like it had significant overhead that could be eliminated by keeping
a few reference counts.

Dave

======================================================================
Dave McCracken          IBM Linux Base Kernel Team      1-512-838-3059
dmccr@us.ibm.com                                        T/L   678-3059

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/