SV1K reliability (summary)

[ian reid]

A long time ago I posted a question on the reliability of the SV1K file
system to the net, and promised a summary.  Unfortunately the response
was underwhelming.  Which is why this summary is so late.  The thread lasted
five or six messages before going off at a noisy tangent discussing what was
and what wasn't a UPS.

In brief the question was why did my system lose files when a power loss
occurred.  I can accept I would lose files which were being modified at the
time, but I was losing (or having corrupted) some indeterminate files
which made my system unbootable.

My configuration was running 386/ix 2.0.1, with a 40Mb disk attatched to a
ST506 controller, inside an Intel301Z box with 9126K extended memory.

I received a reply from Ed Hall <edhall%ives at rand.org> saying :-

I've run ISC's 2.0.1 for over a year, now, with assorted power
failures and crashes (I have a WD-1006SRV controller which locks up
every few weeks or so--a known problem between that model and several
UNIX's).  I've never lost a file except for files which were being
updated during the crash--and even that occurance has been quite rare.

I received a reply from "LCDR Michael E. Dobson" 
<rdc30med at nmrdc1.nmrdc.nnmc.navy.mil> saying:-

My system, an AT&T 3B2/600G running AT&T Sys V R 3.2.2 supposedly has a hardened
file system, however, I have had to restore from a boot floppy and tape on
occaision after a power failure.  Because of this, in addition to an UPS, I
have installed a powerfailure monitor which automaticly begins the shutdown
sequence when it senses a powerfailure from the primary power source.  The
UPS provides sufficient time for users to log off and for the sutdown sequnce to
complete.  For a cost of ~$250 for the transducer/software, it's a very good
investment.  E-mail if you want details on the product.

That was about it.

However around that time my newsfeed went down and I missed some messages
which were quoted as references in Michael's article.  These apparently
according to Michael merely stated why the problem couldn't happen.
They may also have included an explanation of file system hardening which
would still be useful.

Apologies for the (very) late summary.
-- 
Ian Reid 					#include <std/disclaimer.h>
UUCP: ir at cel.uucp or ir at cel.co.uk or    ...!{ukc,mcsun,uunet}!cel!ir
"Computers..proof positive that no-one yet understands how to describe any real
 world situation in 0's and 1's."