5 ms·
> If we look at a worn out drive, one very close to end-of-life, it's specced to retain data for one year to three months, depending on the class of drive. Wha
by Jwarder 7y ago
> If we look at a worn out drive, one very close to end-of-life, it's specced to retain data for one year to three months, depending on the class of drive.
What happens once the data expires? Does the SSD return an error when the data is read, or does it read the bogus data without knowing?
I would kind of prefer the drive bricking itself rather than risking silently backing up bogus data. The earlier comment about ECC and bit error rates suggests bad reads are identified as such, but I'm not sure how far to trust that given that, as mentioned, I/O is hard.
- swiley 7y agoMemory gets corrupted. Harddisks at least would slowly die over time because sectors of the disk would stop working and have to be marked bad. I’m pretty sure with most things the data you read back from memory is a random variable (one with a very very low variance but it’s still random at the end of the day.)
- mastax 7y agoError detection is inherently probabilistic, so there's always the chance that bogus data is read without detection. SSDs use multiple levels of error correction where each level is slower but more reliable than the previous. Such a scheme could only work if the error detection ability of each level were much greater than its error correction ability. I wouldn't rely on SSD firmware to do anything in particular, though. Your best bet is to monitor SMART stats about error rates. If there's a high error rate some of your sectors may be reading bogus data without you being able to detect it.
- tfha 7y agoYou can checksum backup data. Might be more difficult with high speed data like databases, but if it's just backups a cryptographic checksum is really reasonable.