Date: Tue, 5 Mar 96 22:42 EST
   From: Mike Frisch <mfrisch@saturn.tlug.org>

   At 07:37 PM 3/5/96 -0500, David Gilbert wrote:
   >	Well, for total thoughput, I'm sure that you can manufacture
   >numbers that show that the dynamic buffer-cache will be a win.

	   Not just in rare applications, but I find that Linux operates quite
   smoothly on this premise.  Also, the fact that Windows NT does this makes it
   possible to get better overall system performance.  This "dynamic disk
   cache" is certainly something of note and used within reason, I'm sure it
   will be more beneficial than detrimental.

Actually the big win is a split file-page/process-page yet completely
coherent page cache.  Which is what linux has now.  For the
process-page (usually holds text pages) cache you hash the cached
pages as a function of page physical address and the virtual kernel
address of the vfs inode.  It needs to be tuned a lot to get it 'just
right' though especially WRT swapping.

Later,
David S. Miller
davem@caip.rutgers.edu