which file system for sd card?

William Kenworthy billk at iinet.net.au
Mon Jan 26 00:39:17 CET 2009


On Sun, 2009-01-25 at 15:15 +0100, Fernando Martins wrote:
> Pietro "m0nt0" Montorfano wrote:
> > William Kenworthy ha scritto:
> >   
> >> I found vfat clearly better (less susceptible to corruption) than ext2,
> >> however ext3 is better than vfat, but will still play up at times.
> >> Using it for OSM maps
> >>
> >> BillK
> >>     
> >
> > IMHO the best fs for sd is vfat if you don't care about permission, so 
> > if you have to put your photos or docs on it, else ext2 is BETTER than 
> > ext3 in this case for a simple reason, it doesn't have the journal.
> > Not having the journal imlies a redouced number of write to sd and a 
> > reduced probability to get the data lost.
> >
> >   
> I'm not sure I follow your logic favoring ext2. The point of ext3 
> journal is exactly to control for errors. Even if you get globally more 
> chance of errors with ext3, you should only consider the errors in the 
> journal. I mean, a write to ext2 is equivalent to a write in ext3's 
> journal, since from here, ext3 guarantees no errors in the fs, even if 
> takes more tries to update it. So, if the transfer from the journal to 
> the fs itself is guaranteed, then the comparison should be between ext2 
> and the ext3's journal, right?
> 
> Anyway, I'm also inclined for FAT, mostly for the simplicity (thus "less 
> susceptibility to corruption") and universality of the fs. I'm just left 
> wondering about performance, in particular as a storage for maps.
> 
> 
> Cheers,
> Fernando
> 

No, I favour ext3, with ext2 the worst of the lot.  However, its
redundant as vfat wont handle ~840000 map tiles either without
continuing problems (large scale corruption, partitions going read only
part way through data transfers etc).  It doesnt appear very good the
fact that a journal is needed to avoid corruption as a fact of normal
operation.  Ive tried ext2/3 on dirvish server(s) and found a similar
problems - just poor/inadequate/inappropriate choice for this pattern of
data (OSM maps).  Solution, reiserfs with data=journal ...

The consensus was you are very unlikely to kill a modern SD card.  I
also did not see anyone who had actually been able to kill an SD card
from this reason.

BillK







More information about the community mailing list