Re: [Patch 3/3] prepopulate/cache cleared pages

[Paul Jackson <pj@sgi.com>]

Just a random idea, offered with little real understanding
of what's going on ...

  Instead of a per-task clear page, how about a per-cpu clear page,
  or short queue of clear pages?

  This lets the number of clear pages be throttled to whatever
  is worth it.  And it handles such cases as a few threads using
  the clear pages rapidly, while many other threads don't need any,
  with a much higher "average usefulness" per clear page (meaning
  the average time a cleared page sits around wasting memory prior
  to its being used is much shorter.)

  Some locking would still be needed, but per-cpu locking is
  a separate, quicker beast than something like mmap_sem.

Mind you, I am not commenting one way or the other on whether any
of this is a good idea.  Not my expertise ...

-- 
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <pj@sgi.com> 1.925.600.0401