4 ms·

From looking at numbers from the macOS SSD thrashing issue, we know lifetime usage as reported by SMART is not a strict function of terabytes written (at least

From looking at numbers from the macOS SSD thrashing issue, we know lifetime usage as reported by SMART is not a strict function of terabytes written (at least on those SSDs). This makes sense, because small writes have a much higher endurance cost than large writes on SSDs. SSD lifetime isn't based on total bytes written, it's based on total blocks erased in the underlying Flash, and blocks are much larger than sectors so small writes necessarily cause write amplification to some extent (depending on the wear leveling). If manufacturers are guaranteeing a certain number of TBW on SSDs but can't actually sustain that at pathological write sizes, they need to start qualifying those guarantees with e.g. an average write size statistic. Then they can claim Chia breaks that assumption and reject warranty claims in those cases. This isn't unique to Chia. For example, a database workload will cause higher stress per byte written on an SSD than a raw video recording workload. It's also possible they're talking about time-based warranties with daily usage numbers. Many SSDs will be "guaranteed for 5 years" at some DWPD (drive writes per day). In that case they're probably just saying Chia plotting greatly exceeds the DWPD stat and thus your number of years warranty no longer applies. That ignores the above problem and is just about how the drive endurance is specified. (Also, just to clarify: apparently SSDs are used in Chia as a staging ground to build data to put on HDDs, which is then read only; I don't think there is any endurance cost to disks for actually "farming" Chia, only for "plotting" the contents of those disks initially; I think Chia is silly but just want to get the tech right)

Dylan16807 6y ago
Does Chia do lots of small writes? A lot of these plot generation steps sound like the writes are pretty linear. Plus sorting, which can easily be designed to use big writes too.
compressedgas 6y ago
Is there any way to extend the lifetime of a SSD by allowing its capacity to decrease by replenishing its reserve of replacement blocks so that the SSD could continue to operate until it runs out of writable blocks?
- verall 6y ago
  Generally, you would look to a datacenter line of SSD which is rated for much higher endurance. These are typically overprovisioned much more than consumer ssds in order to meet the endurance requirements. I think most OEMs opt not to let you choose this sort of thing and rather bury it into the drive firmware.
- lathiat 6y ago
  Theoretically this would be possible but probably wouldn't help much in practice. Writes on an SSD aren't mapped 1:1 from "logical" address to physical address - they constantly balance and move writes around the entire device (the "Flash Translation Layer" it's sometimes called). As a result once you have exhausted the normal reserve capacity (perhaps allowing for some early life failures) I suspect it's likely the rest of the device will soon follow and you wouldn't extend it's life by that useful an amount. I don't have hard data however on like the "curve" of flash cell failures as an SSD ages and therefor exactly how much longer you might get based on the minimum to median cell failure.
- wtallis 6y ago
  A SSD that presents as a traditional hard-drive style block storage device can't shrink itself on the fly without causing catastrophic software problems. The least bad way to implement such behavior would be for the highest-numbered LBAs to be removed first, which means your backup partition table is first to disappear, then you start encountering filesystem corruption. A zoned SSD has an explicit mechanism for the drive retiring zones from service and making them read-only or completely inaccessible. But it's not currently practical to use a zoned storage device as the primary or sole storage in a system; at a minimum you need a regular block device or namespace to boot an OS from.
- fomine3 6y ago
  https://www.intel.com/content/dam/www/public/us/en/documents/white-papers/over-provisioning-nand-based-ssds-better-endurance-whitepaper.pdf https://www.intel.com/content/dam/www/public/us/en/documents...

Dylan16807 6y ago

Does Chia do lots of small writes? A lot of these plot generation steps sound like the writes are pretty linear. Plus sorting, which can easily be designed to use big writes too.

compressedgas 6y ago

Is there any way to extend the lifetime of a SSD by allowing its capacity to decrease by replenishing its reserve of replacement blocks so that the SSD could continue to operate until it runs out of writable blocks?

verall 6y ago
Generally, you would look to a datacenter line of SSD which is rated for much higher endurance. These are typically overprovisioned much more than consumer ssds in order to meet the endurance requirements. I think most OEMs opt not to let you choose this sort of thing and rather bury it into the drive firmware.
lathiat 6y ago
Theoretically this would be possible but probably wouldn't help much in practice. Writes on an SSD aren't mapped 1:1 from "logical" address to physical address - they constantly balance and move writes around the entire device (the "Flash Translation Layer" it's sometimes called). As a result once you have exhausted the normal reserve capacity (perhaps allowing for some early life failures) I suspect it's likely the rest of the device will soon follow and you wouldn't extend it's life by that useful an amount. I don't have hard data however on like the "curve" of flash cell failures as an SSD ages and therefor exactly how much longer you might get based on the minimum to median cell failure.
wtallis 6y ago
A SSD that presents as a traditional hard-drive style block storage device can't shrink itself on the fly without causing catastrophic software problems. The least bad way to implement such behavior would be for the highest-numbered LBAs to be removed first, which means your backup partition table is first to disappear, then you start encountering filesystem corruption. A zoned SSD has an explicit mechanism for the drive retiring zones from service and making them read-only or completely inaccessible. But it's not currently practical to use a zoned storage device as the primary or sole storage in a system; at a minimum you need a regular block device or namespace to boot an OS from.
fomine3 6y ago
https://www.intel.com/content/dam/www/public/us/en/documents/white-papers/over-provisioning-nand-based-ssds-better-endurance-whitepaper.pdf https://www.intel.com/content/dam/www/public/us/en/documents...