37

It's almost common knowledge that excessive writes (including formatting, defragmenting etc) will, in time, wear out solid state drives. But does reading lots of data from SSD's cause wear too?

I'm planning on enabling prefetching on a Linux machine with a SSD. I have atime disabled.

Wander Nauta
  • 563
  • 1
  • 4
  • 8
  • 3
    I would note that, while it's common knowledge, most people vastly overestimate how fast they'll run out of writes on an SSD. Intel specifically states you can write over 21GB a day for ten years straight without exhausting the writes on their consumer-level drives. – Shinrai Jun 22 '12 at 14:08
  • 1
    (Also, never defragment an SSD anyway, since it won't actually do anything useful.) – Shinrai Jun 22 '12 at 14:08
  • (Also, never defragment an SSD anyway, since it won't actually do anything useful.)   @Shinrai, you mean besides making sure that files are contiguous and thus dramatically increasing the success of data-recovery? – Synetech Aug 19 '13 at 04:41
  • 4
    Think of bending a piece of metal (like a coat-hanger). You can “write” information into it (e.g., straight=0, bent=1) by bending it, and “read” information by looking at it. You can read it as many times as you want without (significant) harm, but you can only write to it so many times before it breaks. – Synetech Aug 19 '13 at 04:44
  • 1
    @Synetech - My impression is that the physical location of data on an SSD is both A: largely irrelevant in 99% of circumstances and B: Not properly understood by most defragmentation software since they're written expecting hard drives, and also the controller on the drive handles most of that stuff (things getting put in different places for wear leveling purposes, etc). Vastly oversimplifying here of course. – Shinrai Aug 19 '13 at 04:46
  • @Shinrai, who’s talking about physical location? The layout of the file is logical and dependent on the file-system. If you delete a file, trying to recover it will require the file to be in one, contiguous block of disk clusters as seen by the OS (regardless of its physical location). Otherwise, there is no way to recover it since the pieces would be scattered around the disk and no way to re-assemble them. SSDs are no different than traditional hard-drives in this regard. With hard-drives, bad sectors are remapped, so a file could be physically fragmented, but logically contiguous. – Synetech Aug 19 '13 at 04:48
  • 5
    Almost all of the information in these comments is out of date. – David Schwartz Dec 11 '15 at 16:31

4 Answers4

40

Doesn't affect the device. The limited write lifetime of Flash is a natural consequence of how they work.

Data on Flash drives is safe because the bits are stored by electrons locked in a very well isolated layer. These electrons, if present, produce an electric field that can be picked up by a nearby transistor. Since they're locked up, reading out the transistor doesn't affect the electrons. During writes, however, to get the electrons through that layer Flash needs very high voltages. These high voltages cause some damage to the isolation layer, which accrues.

In comparison, DRAM doesn't have such an isolation layer. The electrons move quite easily. As a result, DRAM is faster and doesn't break down from writes, but the leaked electrons frequently need to be replaced. Turn off the power and they're all gone in milliseconds.

MSalters
  • 8,165
  • 15
    Although reads themselves are not directly harmful, it is worth noting that excessive reads can cause the firmware to generate background writes. That said, the background writes will probably be insignificant in most cases. More info: http://superuser.com/a/725145/6091 – rob May 08 '14 at 17:10
  • beautiful answer – cedbeu Mar 20 '20 at 02:52
4

I don't believe the read process affects the NAND cells although I could be wrong (for example, look towards the bottom of this article). It may be that if a "page" or eraseblock is not reprogrammed in a very long time there is a (probably very small) likelihood some of the bits will revert to an unprogrammed state. Not sure if firmware takes this into account and rewrites/remaps pages that haven't been read in a long time.

LawrenceC
  • 73,957
  • Interesting. Do you happen to have more information (than just that article) about this behavior? – dtmland Feb 28 '14 at 16:07
  • I don't unfortunately but will try to remember to update if I encounter further info. – LawrenceC Mar 01 '14 at 18:54
  • 2
    It's worth noting that the reads themselves are not harmful, but excessive reads can cause the firmware to generate background writes to counteract retention errors and read errors. @dtmland See my answer to a similar question that was inspired by ultrasawblade's answer. http://superuser.com/a/725145/6091 – rob May 08 '14 at 16:59
3

The reliability section of this table does not mention it, so I assume reads do not affect the drive.

Bart van Heukelom
  • 2,097
  • 3
    That would be my assumption as well, but I would like to be very sure – disabling prefetching is free, new SSDs cost a fortune. – Wander Nauta Jun 22 '12 at 13:18
1

Flash memory is just an eeprom (a chip that can be reprogrammed. It is the reprogramming that causes wear, reads are unlimited. For reading its just memory. http://en.wikipedia.org/wiki/Flash_memory#NAND_flash this article talks a bit about how the reprogramming works, and how it basically 'burns' the data into memory.

Brian
  • 2,982