Reliable storage on an unreliable embedded box?

possibly a small capacitor or battery bank that would supply the system long enough to trigger a safe shutdown? a small ups with a direct dc connection to the board would do it, besides a mains sensor.

--
Using Opera's revolutionary e-mail client: http://www.opera.com/mail/

you can use filesystem like reiserFS or jffs2, but that still doesn't mean that when power loss occurs your block of data will be valid, so I agree with jim, you definetly need some way to maintain power long enough to make sure you don't loose any data.

best regards, Mario

Here we discussed already several times that there is a possibility that a Flash card is destroyed (getting completely unaccessible and needs to be sent to the manufacturer for internal reformatting) when power goes down while a write process is in progress.

The flash manufacturers don't publish any data about ho long power needs to be provided after a write command (that does terminate seemingly correctly, but the card still does a lot of internal work).

But I suppose this usually takes not more than some seconds.

So the only way to securely handle a flash card is to use hardware that provides power for some seconds after the last write to the card has happened.

Of course _additionally_ the Linux system needs to be made write all remaining data from the file system cache to the card before.

-Michael

and it doesn't have to be big and ugly, small and elegant is the way you go if you want efficiency and possibly the ability to squeeze it into the case. no inverter, they suck a lot of power from the battery. a battery sized to fit in a floppy or hard disk bay would fix it fine, if you have a spare.

--
Using Opera's revolutionary e-mail client: http://www.opera.com/mail/

Jim, Mile: Thanks for your advice.

I guess I was hoping that the combination of (1) a journalling FS such as EXT3, (2) mounting the filesystem with options "sync data=journal" and (3) turning off the hdd write cache would ensure that the data is either stored properly or stored in the journal. (ie, the data is not waiting in the OS filesystem cache or the HDD write cache)

I'm not sure that a battery is an ideal solution. The whole box

when the hard-disk is active, will draw up to 15W - and I'll struggle to get a battery to source as much as an amp while the machine shuts down cleanly. :-/

Michael: flash writes

I wasn't intending to ever mount the flash device read-write: this will be a read-only root filesystem.

Matthew

a standard pack of 1400ma nicads that are used by rc racers is about the same length of a hard disk, and half as wide,but also half as thick, can squeeze in about anywhere, and i believe they're 12V. one thing to do is make your own packs, individual cells and tiny buss strips, solder the cells together to make 13V, then use a maxim dc/dc converter chip(something round 95% efficiency). you can of course use a different kind of battery.

--
Using Opera's revolutionary e-mail client: http://www.opera.com/mail/

as

More than 10W of that would be due to the hard disk. Unfortunately, you can't isolate the power for the on-board CF. If you use an external flash disk (is 4G big enough for you?), you can keep it in stand-by for less than 0.1W (approx 0.5W active). CF based flash disk will run in as little as 3.3V and 100 mA. A couple of rechargeable batteries will keep it running for hours.

If you don't want to deal with charging batteries, you can build a simple power isolation circuit with a diode and cap. It works well to extend power to flash disks for seconds or even minuties.

...