file system integrity

[Doug Gwyn]

In article <AZ1RhMO00Ug7M2=JlU at andrew.cmu.edu> sr16+ at andrew.cmu.edu (Seth Benjamin Rothenberg) writes:
>My department will soon be buying a large UNIX box (Vax 5400/5800 or TI
>1000 something).  TI says the file system is secure - i.e., you
>could turn the machine off and on again and no files would be lost,
>and you could log in immediately.  We seem to understand from DEC that
>we would need to run fsck before we could log in, and that this requires
>10 minutes per disk.  We have 12 drives.  We don't have 2 hours to spare.

Generally one runs fsck on all spindles at the same time, which
drastically reduces the wall-clock time for this procedure.

Current releases of UNIX System V are supposed to have fully hardened
file systems, so that after a power outage any on-disk inconsistencies
will not cause problems to spread.  (Of course files being modified at
the time of outage might be corrupted.)  I don't know how true that
really is, or whether you could get away with it using a BSD filesystem.

>Does anyone have any idea what these people are saying?  i.e., did DEC
>write an implementation that doesn't use checkpointing and flush()?

If you shut the system down cleanly, no fsck is necessary.  It's when
the power is suddenly removed that the in-core buffers are not flushed
to disk, causing possible file system corruption.