Re: raid 1 - automatic 'repair' possible?

[Ric Wheeler]

Having looked at a lot of disks, I think that it is definitely worth 
forcing a write to try and invoke the remap.  With large drives, you 
usually several bad sectors in the normal case (drive vendors allocate 
up to a couple thousand spare sectors just for remapping).

Depending on the type of drive error, the act of writing is likely to 
clean the questionable sector and leave you with a perfectly fine disk.

Ric

Jakob Oestergaard wrote:

> On Wed, Jan 19, 2005 at 11:48:52AM +0100, Kiniger wrote:
> ...
>  
>
> > some random thoughts:
> >
> > nowadays hardware sector sizes are much bigger than 512 bytes
> >    
>
> No :)
>
>  
>
> > and
> > the read error may affect some sectors +- the sector which actually
> > returned the error.
> >    
>
> That's right
>
>  
>
> > to keep the handling in userspace as much as possible: 
> >
> > the real problem is the long resync time. therefore it would
> > be sufficient to have a concept of "defective areas" per partition
> > and drive (a few of them, perhaps four or so , would be enough) 
> > which will be excluded from reads/writes and some means to
> > re-synchronize these "defective areas" from the good counterparts
> > of the other disk. This would avoid having the whole partition being
> > marked as defective.
> >    
>
> I wonder if it's really worth it.
>
> The original idea has some merit I think - but what you're suggesting
> here is almost "bad block remapping" with transparent recovery and user
> space policy agents etc. etc.
>
> If a drive has problems reading the platter, it can usually be corrected
> by overwriting the given sector (either the drive can actually overwrite
> the sector in place, or it will re-allocate it with severe read
> performance penalties following). But there's a reason why that sector
> went bad, and you realy want to get the disk replaced.
>
> I think the current policy of marking the disk as failed when it has
> failed is sensible.
>
> Just my 0.02 Euro
>
>  
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/