On Sun, Aug 30, 2009 at 09:33:22PM -0400, Bernie Innocenti wrote:
On a side note, which filesystem should be chosen in order to minimize wearing to USB sticks?
Only the manufacturer can know. Maybe we should ask a few of them. :-|The problem is that the FTL is highly proprietary and very likely tuned to the FAT filesystem (since it's the one mandated by the SD card standard). It's entirely possible that it makes some assumptions that are invalid for non-FAT and does strange things then. This might be causing the abyssmal performance (<< 1MB/s write speed) I experience lately on my XO-1 (with as much structures as possible aligned on 4MB blocks), though having the Journal aligned only to 128KB blocks (ext3) might be part of the problem as well (16GB SiliconPower SDHC card, erase block size unknown). The initial raw write speed was up to 20MB/s (12MB/s on the XO-1), dropping to 4-5MB/s on the XO-1 after installation, now down to 500KB/s max. These figures are for sequential writing of large files (using dd with conv=fsync), random writes seem to be even worse (observed using dstat).
I would assume that the any DOS filesystem will continouously rewrite toThe increase of data written to the card with journalling enabled is less than 4% for normal operations (up to 42% for very metadata-intensive operations, but then the amount written is low in absolute terms) according to Ted Ts'o [1] (who has quite a number of interesting SSD-related articles in his blog). For me having a journalling filesystem is worth a lot more than just 4% write data increase (I even use data=journal on most systems).the FAT blocks. Maybe the best choice would be ext4 with the journal disabled?
[1] http://thunk.org/tytso/blog/2009/03/01/ssds-journaling-and-noatimerelatime/
CU Sascha -- http://sascha.silbe.org/ http://www.infra-silbe.de/
signature.asc
Description: Digital signature
_______________________________________________ Sugar-devel mailing list Sugar-devel@lists.sugarlabs.org http://lists.sugarlabs.org/listinfo/sugar-devel
