The new Sun Enterprise Server 450, (called sassrv.ilo.ch), has about 18Gb of free disk space. The estimated total size of the STAT files that must be migrated from MVS is about 250Mb, hence there is no present need to be concerned about disk space limitations. If the activities of STAT do not change significantly in the future and not too many departments share the machine, there may never be a need to worry about space limitations.
Presently, backups of the complete server file system to
12Gb-uncompressed/24Gb-compressed tape cartridges are done nightly
(around midnight) using ufsdump(1)
. ufsdump(1)
is a
simple program for incremental file system backup that requires the
source file system to be inactive during the save process. The
process takes approximately twenty minutes. If there is ever a need to
have 24/7 availability of the server, a more elaborate solution will
be required.
The present system is cumbersome for retrieving specific backed-up files since it requires manual operator assistance to load the correct backup cartridge. Two companies have already proposed automated solutions for managing banks of cartridges.
Although Solaris 7 ships with a backup program, the licence is only for a thirty day demo. There is very little information on USENET about its reliability and there are hints in the accompanying documentation that Sun is thinking of discontinuing support for this product.
For the moment, we are contemplating migrating the files from MVS to Unix and keeping their file names, and relative positions intact. Letter-case of filenames will not be changed immediately and the only translation of MVS names to Unix names will be the translation of the periods "." to slashes "/". Since the names of these files are mostly transparent to SAS users, no users will notice if modifications of directory structures and letter-case are made in the future.
For the instant, the priority is to get a working system going on Unix and the fewer structural modifications that are made, the less likely a number of scripts will break.
I believe that the STAT administrators will find it easier to debug problems caused by the migration if they don't have to get used to a completely different directory structure simultaneously. This policy will be reconsidered once the administrators feel that they will be more comfortable with a different directory layout and when they are fed-up with toggling the CAPS-LOCK key in order to specify old MVS filenames.
A 8.5Gb partition mounted at the /data
directory has been
created for storing SAS databases and related files. This directory
will eventually be shared with other departments, so a subdirectory
for STAT-only files has been created called stat
.
Under the directory /data/stat
there is a staging directory
called test
and a production directory called prod
.
Program development should take place in each developer's home
directory /home/
username if there is more than one
developer working at a time. If it can be garanteed that no
interference will occur, it is acceptable that development take place
directly in /data/stat/test
. A CVS (control version system)
would be over-kill for STAT.
All of the files from MVS will be stored under
/data/stat/test
during the December testing phase with their
MVS dataset names intact. Hence, a sequential dataset on MVS that was
called LOSTPC.TRANSF.ZIP
will be found on the Sun at
/data/stat/test/LOSTPC/TRANSF/ZIP
and a PDS member on MVS
called LO.ST.CRDX.ILOSTAT(FORMATS)
will be found on the Sun
at /data/stat/test/LO/ST/CRDX/ILOSTAT/FORMATS
.
When the STAT administrators are finally comfortable with the new Unix system, I suggest that the following directory structure be adopted. It roughly follows traditional Unix directory hierachy, and is somewhat easier to use since several levels of directories are removed.
/data/stat/test/LO*
subdirectories to a new directory called
/data/stat/test/sas
.
/data/stat/test/LO/ST
up to
/data/stat/test
. This is done with the command
mv /data/stat/test/LO/ST/* /data/stat/test
/data/stat/bin
and use it to store
Unix programs and scripts that everyone in STAT should have access to.
Examples of what should go in this directory include all of the
scripts written for this migration project, the main STAT6 menu
system, and the translated JCL scripts for creating monthly bulletins.
/home/
username directories and moving programs to
the public shared directory only when they are done. SAS users should
never find half-finished or modified programs anywhere under
/data/stat
, only pure data and working programs.
/data/stat
to lower
case. This can be easily done with a shell script.
/data/stat/test
tree to /data/stat/prod
and set the
STAT users' environments accordingly.Tampering of the SAS database files was prevented on the MVS system by
a home-made script (a "CLIST") that checked to see if one's user ID
was on a list of user IDs allowed to modify certain files. Each file
had it's own list of authorised user IDs. The lists were written
directly into the script. When the script was migrated, this
home-made security system was kept intact, but the access lists were
migrated to an easy-to-edit configuration file called
/data/stat/test/etc/stat6access.conf
.
A better (or at least more "Unix-like") solution is to set different group protections on each data file and then give each user appropriate group permissions. Although the old system was sufficiently capable of preventing honest mistakes, it was decided that a group-protection system should be implemented.
SAS (like many other database systems) has its own file-locking system
to permit simultaneous access to datafiles. This has been purchased,
but not yet installed. Until it is installed and configured
correctly, there is a risk that two people working with the same data
file may over-write each others' work. Solaris 7 also has
file-locking capabilities (set with chmod(1)
), but since the
SAS system will probably be more suitable, no effort was made to use
it.