rhkramer@gmail.com writes:
On Wednesday, January 11, 2023 12:20:05 PM Linux-Fan wrote: > > Or does one need to read every byte, allocated or not? > > AFAIK one needs to _power_ the device every once in a while and keep power > connected for some time. Then, the controller can dos all the necessary > actions in the background.
[...]
> This entry seems to be rather pessimistic: > https://www.ibm.com/support/pages/potential-ssd-data-loss-after-extended-sh > utdown I've read some of that article, or, I guess, really the abstract and the section labeled "Content" on that page: https://www.ibm.com/support/pages/potential-ssd-data-loss-after-extended- shutdown I see the statement: "A system (and its enclosed drives) should be powered upat least 2 weeks after 2 months of system power off. If a drive has an errorindicating it is at end of life we recommend not powering off the system for extended periods of time.", and the first quoted paragraph in this emailreiterates the need to power up occasionally and leave connected for some time(so that the controller can do all the necessary actions in the background). I assume that they are talking about the hardwired controller built into the drive, thus there is no particular need to power it up from an OS that recognizes it but simply something that powers it somehow?
Actually, since the IBM article is about some sort of storage system, it becomes hard to tell what IBM mean with the "controller": The two weeks are quite long a period i.e. they expect some technician to swap SSDs in the storage array once in a while and in there, they are powered 24/7.
For consumer SSDs, I understand the controller to be indeed one of the chips on the SSD i.e. outside the realm of the OSes control. Also, no consumer SSD manufacturer advises their customers to power the SSDs two weeks straight :)
Can anyone shed more light on what happens during that two weeks -- is thedata somehow "refreshed" (in place), or rewritten somewhere else on the drive, or ???(Perhaps that is discussed in the complete article of which this appears to be just the abstract??)
For the IBM case I expect that information to be proprietary :)In fact, there seems to be curiously little information available on the topic at all. Most people (including credible sources like e.g. the Kingston Support) suggest that all data should be read from the flash for the controller to refresh it:
https://reboot.pro/index.php?s=234d281f8a9f18ba7b36f5e98890bd2f&showtopic=13791#entry121995It is probably not the entire truth though, because if I understand some JEDEC slides correcly, thery only talk about power-off where I assume that everything non-power-off must not count towards the data retention time including not reading the data?
(cf. slides 26f of https://www.jedec.org/sites/default/files/Alvin_Cox%20[Compatibility%20Mode]_0.pdf) Official manufacturer documents are pretty opaque regarding the issue: https://www.kingston.com/en/blog/pc-performance/ssd-garbage-collection-trim-explained https://semiconductor.samsung.com/resources/white-paper/Samsung_SSD_White_Paper.pdf (relevant chapter is CH04 with pp. 17ff) [...]Maybe someone with advanced Interenet searching capabilities can bring up more relevant documents about the subject :)
HTH Linux-Fan öö
Attachment:
pgpJQ9dyCuUlj.pgp
Description: PGP signature