ESIX vs. SCO Xenix

[Piercarlo Grandi]

In article <228 at bradf.UUCP> brad at bradf.UUCP (Bradley W. Fisher) writes:

   There is one aspect of the *NIX wars that I haven't seen discussed
   yet ... one that I consider to be of the utmost importance ...
   how long does it take to back/restore from tape? So far comparing 
   SCO, Altos, NCR, and Interacive , SCO wins hands down! Typically, 
   80 megabytes will stream to tape in about 40 minutes under SCO with
   a standard Wangtek 5099EN tape drive and PC-36 controller. The same 
   amount of data and tape system takes about 3 hours under Interactive.
   Altos and NCR are slightly faster than 386ix, but *nobody* beats SCO
   in this respect from what I've seen. Why is this?

The reason is that you are not using any of them properly, and SCO uses a dirty
and potentially dangerous trick. In order to make the tape stream, you
have to overlap disc reads with tape writes (or viceversa). When the
tape streams, it will take less than 20 Minutes to dump 80 megs, not 40
as you are getting.

To overlap disc and tape IO you need buffering and two processes or
more. Try using my own 'team', posted to alt.sources, or 'ddd', posted
to comp.sources.misc, as tape writing filters. Also, try using GNU tar
or pax or afio instead of dump; they are more portable and often faster,
even if admittedly dump/restor is faster on restore...

The trick that SCO uses is to make the tape driver grab a tape buffer
from the filesystem buffer cache, and then write this asynchronously to
the tape device. This sort of works, but does not provide sufficient
overlap, and completely alters the semantics of raw device writes.

The bottom line is that with the right tools you get 4-5 megs per minute
on any UNIX system that has enough filesystem read bandwidth and a QIC tape.
--
Piercarlo "Peter" Grandi           | ARPA: pcg%cs.aber.ac.uk at nsfnet-relay.ac.uk
Dept of CS, UCW Aberystwyth        | UUCP: ...!mcvax!ukc!aber-cs!pcg
Penglais, Aberystwyth SY23 3BZ, UK | INET: pcg at cs.aber.ac.uk