I never really saw the point in ext3, and for this application the benefits are even less clear.

ext3 doesn't give any differences (improvements) over ext2 in read-only mode, which is the primary mode of the player. And it complicates the write path significantly while only giving doubtful benefits.

It creates more memory pressure, because it needs to keep data writes and meta-data updates 'ordered', not too helpful on a machine with only 12MB of memory. Especially when the ext3 extensions and the jbd layer have already added more unswappable kernel memory to the kernel image.

In data=ordered journalling mode it increases the amount of head seeks significantly because it has to sync all the data blocks before it writes the metadata updates to the journal, and then still has the update the actual ext2 inodes and bitmaps.

Data journalled does reduce seeks but it ends up being even worse, especially on an empeg, as it writes the data to both the journal and it's final location. And the bottleneck on the empeg is the relatively slow cpu->disk path which uses PIO instead of DMA.

The third mode that ext3 provides is useless as it only guarantees metadata consistency. It actually hides problems and lost data that fsck could normally detect.

And all of this to save an occasional fsck that in most cases isn't even needed and that in all the years of using Linux has only failed on me a few times. And that was because I was running a development kernel with buggy drivers that were writing blocks to the wrong location on disk, definitely not fsck or ext2's fault.

I guess I just don't get it.