Re: Fw: [Bug 4520] New: /proc/*/maps fragments too quickly compared to

[Wolfgang Wander <wwc@rentec.com>]

Andrew Morton writes:
 > "Chen, Kenneth W" <kenneth.w.chen@intel.com> wrote:
 > >
 > > Andrew Morton wrote Tuesday, May 10, 2005 11:58 AM
 > > > "Chen, Kenneth W" <kenneth.w.chen@intel.com> wrote:
 > > > > Andrew Morton wrote on Monday, May 09, 2005 2:27 PM
 > > > > > Possibly for the 2.6.12 release the safest approach would be to just
 > > > > > disable the free area cache while we think about it.
 > > > > 
 > > > > I hope people are not thinking permanently kill the free area cache
 > > > > algorithm.  It is known to give a large percentage of performance gain
 > > > > on specweb SSL benchmark.  I think it gives 4-5% gain from free area
 > > > > cache algorithm.
 > > > 
 > > > It also makes previously-working workloads completely *fail*.
 > > 
 > > I agree that functionality over rule most of everything else.  Though, I
 > > do want to bring to your attention on how much performance regression we
 > > will see if the free area cache is completely disabled.  I rather make
 > > noise now instead of a couple month down the road :-)
 > 
 > Well we allegedly have a patch from Wolfgang which fixes things up, but our
 > talk-to-testing ratio seems to be infinite.
 > 
 > This is pretty serious, guys.  Could someone please find the time to work
 > on it?

I volunteer to do the testing - just the test I got from Ingo did
not show any timing difference for either of the three solutions:

a) use free_cache
b) disable free_cache
c) use my maybe improved and maybe much too complex free_cache

The test_str02.c I got only ran up to 1300 threads on my machine (8GB
dual x86_64) and Ingo expected it to go up to 20000.

If there is any other test case I'm very willing to do the timing
tests...

             Wolfgang
--
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/ .
Don't email: <a href=mailto:"aart@kvack.org"> aart@kvack.org </a>