From:	MERC::"uunet!CRVAX.SRI.COM!RELAY-INFO-VAX" 19-JAN-1993 03:15:54.57
To:	uunet!"info-vax@crvax.sri.com"
CC:	
Subj:	More remote disk speculation

Just as a followon to the post I did a few days ago about remote
disks, let me suggest the following:
  If you want a remote disk that is r/w at many sites, one way to
do it, assuming all ods-2 disks and vms machines, is to use something
like my fddriver based one to start. Then you arrange to capture
XQP calls, either by stealing a vector to the XQP entry itself
so you get control ahead of the XQP at the normal entry, or by
adding FDT code to get the info. You transmit open requests etc.
to the other side, but allow open and so on to happen locally also.
The server has to open files for read also that you open. The
trick is that on create/delete/extend, the server must do these
operations. On create/extend you just let the server do it,
then do a local open on the file (which will read it across the
net). On extend, invalidate the window for the file and let
it be rebuilt after the server has extended. Delete can just be
done remotely and not locally. The result is that the local system
(not connected direct to the disk) sees an apparently valid
disk, but the side that has the disk really there manages the
storage. Actual read/write code just uses the direct to disk
path, so one need not fake all the r/w virtual calls, and the
thing will on the whole not perform too badly.
   Anybody got any thoughts on this? Suppose DEC did DFS this
way?
Glenn Everhart
Everhart@raxco.com