On Nov 30, 12:02am, Jonathan Stone wrote:
> but does config-time tuning make sense?  If the 1024 in the old-Mach
> VM really refers to PAGER_MAP_SIZE bytes (4MB or 1k pagesin vm_pager.c),
> there seems no point in allocating more PTEs.

  The 1024 does still refer to the PAGER_MAP_SIZE in vm_pager.c, so it
should stay fixed.

  Chris Jones had a problem earlier this year with not enough PTE entries
on a machine with lots of memory and lots of users, but I think changing
that 256 fixed size to a value based on maxproc may take care of this.  I
could try to have the sysadmin on that particular system try this.  [The
workaround on that system currently just doubles Sysmapsize.]

> Michael is the 2048 based on a UVM parameter, or is it just
> `more than 1k'?

  'more than 1k'

  There are a number of different segments allocated out of the kernel
virtual address space besides the data for the swapper.  I haven't tracked
down each and every on of them yet, but I did find a couple of allocations
done by kmeminit(), which are based upon the size of VM_KMEM_SIZE.  [I.e.,
it allocates a couple of areas, kmemusage and kmem_map, that are in outside
the VM_KMEM_SIZE memory.]

Michael

-- 
Michael L. Hitch			mhitch@montana.edu
Computer Consultant
Information Technology Center
Montana State University	Bozeman, MT	USA