More on hanging tapes under Unix

[NEP.FOUTS%ames-vmsb at sri-unix.UUCP]

     In an environment where users are not allowed physical access to the
machines, for whatever reason (in addition to security / paranoia on the
part of those who operate the machines, there is also the case of the
remote user,) there must be an ability to instruct the operator using
simple command level procedures, with predefined protocals.  After all,
the operator cannot always be expected to know how your latest software
wants him to respond to a tape hanging request.

     The 'hang' command discussed on the net is an excelent starting point
towards bringing UNIX towards production environment use of tape drives.

     An additional form of tape hanging in an environment where "real"
operators cover all three shifts of operations is to make tape hanging as
invisible as possible to the user.  There are two cases where this can be
applied relatively easily.

     First, it is possible to design a hierarchial file migration system
which causes files to migrate from fast, expensive disk to slow, cheap
tape, to archive.  In this case, the file entry still exists in user
directory, with a pointer to the actual storage medium.  When the user
wants the file again, the operating system invokes routines which cause
the operator to hang the correct tape, copies the file down to the disk
and then turns it over to the user.

     Since the time to retrieve a file is nontrivial, there has to be a
mechanism to request a file before it is needed, along with a way to
gracefully recover if the user doesn't want to wait for the file.

     Second, it is possible to define a directory hierachy such that the
medium of storage is not readily apparent.  This is typically done in a
similar fashion, with directory entries on line pointing to files on tape.

     To use this mechanism, the file creator is given the option of specifying
the medium, and the medium is transparent to all other users.  Again, the
retrieval time is nontrivial, and an abort option must be available.

     The next obvious extension to this is to make not only the medium
transparent, but the actual machine location.  This would be useful in the
case of a large research installation where a Cray with a small number of
local disks was supported by a mass storage system, possibly blue.  It would
make life much easier for both the system user and the system designers if
the software didn't have to know where in the hardware topology a file resided,
but rather only where in a logical file system it could be found.
------