K 10
svn:author
V 5
markj
K 8
svn:date
V 27
2016-07-20T00:43:26.407857Z
K 7
svn:log
V 888
Rework shortfall laundering.

- Redefine the shortfall threshold using the vm_laundering_needed()
  predicate. Now, we are in shortfall when the inactive queue is below
  its target and the free page count is below the pagedaemon wakeup
  threshold. In this state we attempt to launder enough pages to meet
  the inactive and free page targets. This improves behaviour in the
  case where almost all of the system's non-wired memory is active, but
  there is little memory pressure. In particular, there's no need to
  launder aggressively if v_inact_count < v_inact_target and
  v_free_count is above the pagedaemon wakeup threshold.
- The performance is unchanged when the system is persistently in
  shortfall, for example when many threads are writing sequentially to
  large memory-mapped files.
- Add some comments which hopefully make the logic easier to follow.

Reviewed by:	alc

END