On Fri, Mar 26, 2010 at 12:29:57PM +0100, Karsten König wrote: > I'll awnser all in one here > > harddrive is a no go, don't want it running all the time, it's only for > backups > > the reduced writes (no atime, ext2, etc) are already active though I need to > set a higher "block io" time so it will gather more data to write at once > > the card was a class 10 4gb sd card from sansa, as it cost 25 € i thought > it'd > be decent =/
Well if they assume (which they shouldn't but do) that it will be used in a digital camera then they can assume it will use FAT filesystem and will be storing large files. This means they can assume the start of the card where the FAT is stored will be written lots, and the rest of the card won't. In that case they can implement wear leveling that only worries about the first bit of the card and ignores the rest. If you then go and run ext2 or ext3 or some other modern filesystem on it, now all of a sudden the filesystem metadata is spread around, and the wear leveling isn't being done in the right area. A good design would simply wear level the whole card properly and then you could do anything you want with the card (even run swap space if desired). Where I work we use XceedCF from Smart Modular Technologies. We have never had a card fail yet, and we run ext2 and ext3 partitions on the cards. They are fast (I have done 30MB/sec read in UDMA IDE mode on them), and have good wear leveling design. Unfortunately I don't believe they are sold anywhere retail, only through distributers. They are industrial grade cards. > I can't see that from the outside right? Is the internal flash slc? Most likely you wouldn't know. It could even change over time for a given type of card. Many companies don't think it matters to the customer how they implement things as long as the feature set they promise doesn't change. > Yes, I formatted it and copied the old system over, took ages, especially > unmounting took multiple minutes so I guess the write queue was still packed > full and hard to really get onto the sd SD cards unfortunately state that the filesystem will be FAT right in the specifications. So unlike compact flash, SD card makers actually have a justified excuse to implement the wear leveling in a shitty manner that only works well for FAT filesystems. The SD spec states what the filesystem will be and that there won't be more than one partition. If you do anything else, you are on your own. > I guess I just hit a really bad sd card, I'll attempt another go with using > ubifs and the nand flash of the sheeva plug Some cards are of course crap. In general cards that are rated for industrial use are much longer life (often 1000000 cycles per cell versus 10000 was common a few years ago). -- Len Sorensen -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected] Archive: http://lists.debian.org/[email protected]
