Filesystem errors

[peter nordlander]

We have problems with the root filesystem on an IBM RS6000/320.
We keep loosing files on /etc and /lib. The loss appears to be
random. The problem does not appear immediately after system
installation but when users begin to log in and use the computer.

If we reboot after the problem has occurred, the automatic fsck
discovers filesystem errors: Bad Inode Map(NOT SALVAGED),
Filesystem integrity is not guaranteed. These errors occur for
the root filesystem.  After these messages, the fsck is aborted
and the system comes up.  A manual fsck on root now does not
detect any problems. No files are placed in lost+found.  ncheck
however reveals big problems with the inode map. System
engineers from IBM have run diagnostics on the disks and disc
controller but no hardware errors have been detected.  We have
also installed the system on external discs but the problem
always reappear. IBM software support have been working on
this problem for more than a week but are at present unable
to solve the problem.  We have tried different versions of AIX
(03.01.002 and 03.01.003) but the problem eventually reappears. 
  
We have a vague suspicion that the loss of files is larger
during heavy load and frequent swapping. We have 32MB memory and
64Mb swap space( We have increased the swap space but this did
not help). A couple of times we assigned swapping to a separate
disk to make sure that it was not the paging that damaged the
files.  This did not help either. 

We would welcome information from anyone
having similar experiences or any
ideas about how to solve this problem.

Peter Nordlander
Assistant Professor
Department of Physics
Rice University
Houston, TX 77251