[Arm-netbook] Logging and journaling
Tzafrir Cohen
tzafrir at cohens.org.il
Thu Feb 9 10:22:43 GMT 2017
On Wed, Feb 08, 2017 at 11:20:32PM -0500, Julie Marchant wrote:
> On 02/08/2017 07:15 PM, Julie Marchant wrote:
> > Mike, if that article is accurate, the study doesn't contradict what I
> > said:
> >
> >> None of the drives in the study came anywhere near their write limits,
> >> even the 3,000 writes specified for the MLC drives
>
> I wrote this while I was at work on my break, so now that I have more
> time, let me elaborate.
>
> Supposedly, in all cases tested, none of the SSDs were written to
> anywhere near enough times to cause substantial wear. It is mentioned in
> this article summarizing it that even 3,000 writes were not reached.
> Most flash media can sustain at least tens of thousands of writes, so if
> not even 3,000 writes were reached, it makes sense that none of the SSDs
> tested failed.
>
> However, it doesn't follow that this kind of wear is insignificant for
> all flash memory. In particular:
>
> * SSDs tend to be larger than SD cards, so they're not going to wear out
> as fast unless the entirety of the space is being used and constantly
> changed.
> * The firmware of SSDs may be different from the firmware of SD cards.
> For that matter, the firmware of SSDs may be different from the firmware
> of other SSDs. Don't underestimate the power of wear leveling.
>
> It still holds true that every change to any flash memory brings it
> closer to the end of its life. It's just that other factors can do this,
> too, and in some cases (e.g. SSDs) it may mean that the write cycle
> limit is less important in practice.
A test we did for our systems is to constantly (re)write and sync a large
random file for many cycles (enough to simulate several years of normal
operation), and generally encountered hardly any faults.
Feel free to test this yourself.
--
Tzafrir Cohen | tzafrir at jabber.org | VIM is
http://tzafrir.org.il | | a Mutt's
tzafrir at cohens.org.il | | best
tzafrir at debian.org | | friend
More information about the arm-netbook
mailing list