hi,

I made some changes last weekend that were supposed to get rid of
all of these problems entirely for the default configuration.
it works pretty well for me, but I'm guessing that it's not working
as intended for you.

is process anonymous data getting paged out for you when it shouldn't now?
the default configuration is that the system will allow anonymous data
and program executables together to use up to 50% of memory before any
of it will be paged out.  is that not happening?  note that idle processes
will still be "swapped out" even with the new changes, but their data
and executable pages will still be in memory even though ps won't show
it anymore.  these days, "swapped out" just means that some of the
kernel data structures that describe a process are reclaimed, not that
the memory holding the process's data itself is reclaimed.  if you
want to get a better picture of what's going on, insert this line at
the top of uvm_scheduler():

	for (;;) tsleep(&proc0, PVM, "noswap", 0);

-Chuck


On Tue, Dec 11, 2001 at 10:00:05PM -0500, Laine Stump wrote:
> Is it just my imagination, or have others seen it too?
> 
> Yesterday I updated from binaries based on a Nov 24 cvs update to
> source based on Dec 10, and now my system is once again *very*
> unresponsive when I'm (for example), doing a tar of a directory tree
> into a 450MB tarfile (no compression).
> 
> This is the way things were until a couple months ago, when they
> straightened out quite nicely, maybe even a bit worse. Before, the
> only problem was that emacs would become *completely* swapped out and
> take forever to swap back in; now even my mouse pointer is several
> seconds behind mouse movements (I'm running VNC as an X server,
> displayed remotely on a Windows PC).
> 
> Has something changed, or am I just a hypochondriac?