On 11/01/2012 11:48 AM, Neil wrote:
> On 10/31/2012 5:57 AM, alan.b.pearce@stfc.ac.uk wrote:
>> This must come down to how much RAM you have available and how often
>> you are writing a line of data. If writing a line every minute (say)
>> to log temperate and pressure, then writing to the card on every line
>> would be OK, as the card will then deal with a sectors worth of data
>> write to the Flash, and it won't take a minute to do that, so it will
>> be ready in time for the next line write. But if the speed of your
>> writing means that you are writing a line fast enough that the next
>> line write attempt will occur before the card has a chance to write a
>> sectors worth to Flash, then buffering would be a better idea - until
>> you are wanting to write a sectors worth faster than the card can
>> handle ...
> I'm writing approx 60 bytes of data per "transaction" (one transaction
> per line), and approx 20 transactions per second. That's 1kB/s, far
> within the card's spec.  I've not yet found a way to tell if the card
> has completed a transaction, before sending the next.  If/when I do, I
> can run tests to verify it's written and buffer if not.  Is there any
> reason to suspect that the card may not immediately write a transaction,
> and I would have to verify each operation is complete?  I'm using SPI
> mode btw.
>
> I'm going off the info / docs / examples provided with the development
> tools, so I suspect I should find the SD-card SPI spec.
>
>
That's a high write rate.  Is power rock solid, or does your solution=20
allow you to detect power loss early enough to complete writes before=20
power is gone?

We had/have a serious problem with SD cards in systems with unreliable=20
power.  We'd end up with corrupted cards pretty quickly... and we were=20
writing less frequently (maybe 300 bytes twice a second normally).

One big problem is the wear levelling.  ANY sector on the disk can be=20
lost at any power cycle.  There's no amount of write verification you=20
can do to be sure you're safe.  With modern filesystems, writes are=20
always going on at the same file structure portion of the disk... so=20
many writes are going to cause a wear level operation and swap with some=20
other random part of the drive.

There are SD cards that are supposedly power failure tolerant. We're=20
going to look into those a bit more, but we haven't used any yet.

We considered going raw disk and writing incremental data as you would=20
normally do for fault tolerance in FLASH, but it became obvious the SD=20
cards erase the sector first even if they technically don't have to.

We ended up caching one day's data to an internal NAND-flash drive with=20
proper power failure handling, and we quickly dump this data to SD=20
exactly once a day to limit our exposure to the SD card power failure=20
problem.  Writing 172,800 times less often should take our MTBF due to=20
this specific problem from 1 month to... well... somewhere in the=20
thousands of years.  (Depends on write durations, OS behavior, etc)   I=20
still want to find tolerant SD cards to completely get rid of this issue.


Darron

--=20
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist
.