> Wouldn't it be much simpler to just change SPXTAPE to work with
> SPOL-format disk volumes in addition to tapes?  
> Then you could backup to
> dasd and then DDR to tape at your convenience.  Or, just add the
volume to
> the CP-owned list and it's a miracle: the spool files are on it are
now in
> the system.  Oh, and get SPXTAPE to support virtual input/output
devices
> so you could use minidisks.  All of your existing image-backup
functions
> would work against spool files without having to use SPXTAPE-to-tape
and
> you wouldn't have to worry about copying the "live" packs or having
> mindisk overlays of spool volumes.  (It even enables us to have the
system
> *prevent* a minidisk overlay.)

It'd be a lot simpler and smarter long term to get SPXTAPE out of the
business of doing direct device I/O entirely. That's a function that is
trivially manageable in userspace; doesn't need to be in CP, and just
complicates the maintenance of the service over time. See earlier
suggestion wrt a system service, and then just connect the output of the
pipeline (ex "SPDUMP STREAM ALL | TAP1" or "SPDUMP STREAM ALL | > fn ft
fm", etc. Restores could look like "PIPE < fn ft fm | SPREST STREAM" 

All the selection and spool management code in SPXTAPE would be
salvageable from the current implementation; you'd need a userspace exec
to connect up the pipelines and the implementation of the IUCV service
to present the data. No future mods for new device types, and a lot less
testing needed.

> For high-tech backups, I'd tweak the *SPL system service to work with
SDFs
> such that the guest would be notified when there's a new SDF, just
like
> it's notified of new spool files that have specific DEST values.  It
could
> then read the SDF using the same mechanisms and do whatever it wanted.
All
> of the spool metadata would be available.  This infrastructure let's
you
> perform immediate archives whenever something changes (e.g. an updated
> DCSS) rather than waiting for the nightly backup.  Naturally a new
WRITE
> function would be needed to allow the SDF to be restored.

Interesting, but orthogonal to the problem, I think. I think that'd be a
2nd order gadget, once you've fixed the more basic problem of SPXTAPE
having to deal directly with devices. 

Reply via email to