SI Reasoning <[EMAIL PROTECTED]> writes:
> > about "reiserfs ate my /var partition", note that
> > ext3 writeback mode where
> > metadata only are journalized, you can still have
> > old date or mix of old and new
> > data in your files on remount after crash.
> > ext3 ordered mode should be safer. but :
> >
> > note also that for eide disks, most of them works in
> > writeback mode by default,
> > ie the eide controller lye to the OS and say "ok, it
> > has just been written on
> > the disk" whereas it's only in the disk cache.
> > so eide disks can lead to corruption dispite the
> > medata journal.
> > scsi disks should be safer here.
> >
> > there're also other hw problems than can lead to
> > metadata corruption like the
> > via bugs, the dma engine still writing do disk when
> > the power is shoot down
> > because it's powered longer than ram (ram needs to
> > be refresh thousands times
> > per second to keep its data), ...
>
> does the eide disk problem also affect reiserfs?
eide writeback mode affects all fs : by saying "ok it's writing" whereas it's
only cached, the eide disk/controller couple waits for mode I/O to do them more
intelligently if they can order them, they can achieve better write performance
but if a power problem or a kernel oops occurs, then you can lost your data
since they're not yet written.
this can break journaled fs since they rely on physically written metadata.
but the odds are high you see data corruption before metadata corruption.
> Should I disable writeback mode?
safer but slower on writes :-)
if you want, put a hdparm -W 0 /dev/hd[XYZ] in your /etc/rc.d/rc.local
