At the risk of mixing this pot up more, I think I understand
the confusion that is going on here in AFS land about SGI and
sparse files.

FYI, Masscomp (aka Concurrent), was the first to support ``extent
based file systems'' for UNIX -- with Stellar and SGI also later
supporting them (in 1981).  These are also called ``real-time'' or
``contiguous files'' because they lay the file down in contiguous
areas of the disks.  They are different from traditional UNIX file
systems, in that perform better (that is usually reasonably easy)
and predictably (very hard).

While Masscomp did not allow true extents for all files (like Stellar
and SGI), sparse files are not allowed in contiguous files of any
of the extent based systems.  The reason is that during rewrite of
the file, if you allowed sparse layout, you would lose the performance
and >>predicable<< behavior needed and required by those systems.

Anyway, in the mid-1980's SGI wrote their ``EFS'' - extent based
file system.  I believe there is a USENIX paper, althought the
paper I read years ago may have been an SGI ``white-paper'' and never
published.  Thus sparse files went away for SGI with the introduction
of an extent base FS.

Other UNIX systems which still use the V7 or BSD (aka UFS) file
systems offer sparse files (which save on-disk space).  I'm not
sure what Veratas does in this area?  Anybody know?

Clem Cole
Sr. Scientist
Locus Computing Corp

Reply via email to