Re: [opensuse] SSD in openSUSE.



On Wed, Nov 23, 2011 at 15:57, <smartysmart34@xxxxxx> wrote:
On Wed, Nov 23, 2011 at 09:16, Hans Witvliet <suse@xxxxxxxxxxx> wrote:

Why is everyone still perpetuating the belief that Flash based SSDs
still need this careful management of write cycles? Do the math...
with current SSDs and their write cycle ratings (a current/new SSD
does not have a write cycle rating of 10,000 like they did 4 or 5
years ago... you are looking at 5,000,000 or more write cycles now),
you would have to basically do many hundreds of GBs (or even several
TB on larger drives) of writing every single day for 40 to 50 years or
even more to hit the write access barriers. Everyone is basing this
whole "I have to minimize write cycles" thing on SSD specs of several
years ago.

I do not know where you got this info from but the opposite is the case. With shrinking structure sizes to around 25 nm (and less in the near future) the amount of write cycles dropped drastically to approximately 3.000 to 5.000 write cycles.
The manufacturers often give a number called Total Bytes Written and for a current Crucial M4 this is somewhere around 72 TB.
This represents 40 GB/day over 5 years.
For a normal system this may still be plenty, BUT: If you have a SWAP-Partition or the pagefile on that SSD and are hibernating a 24 GB Workstation twice a day this may surely affect your SSD at some point.


Then color me totally confused. There is SO much conflicting
information. The vendors state one thing, people on blogs, forums and
mailing lists state another, and none of the info stitches together
neatly. I got my info from hours of digging through PDF files, spec
sheets, presentations etc. Almost all of it pointed at larger not
smaller write cycles in new SSD devices. I've seen a couple articles
about the Crucial M4 recently (today) with the 3-5000 write cycle, but
that does not corroborate with the 1-5 million write cycles I see
elsewhere.

Also, look at the specs of the Intel 710... how do those specs relate?
(20% over provisioning, 2million hrs MTBF, 1PB of 8k writes over
lifetime.. the 720 is claimed to have 36PB).

Look at the data here
http://www.tomshardware.com/reviews/ssd-reliability-failure-rate,2923.html
Which interestingly states 3000 - 5000 writes... even at that, doing
continuous writes, you're looking at (as you pointed out) 20 to 60
years (depending on the model) of sustained writes at an average of
10GB/day.

Regardless of 5000 or 5000000 (depending on who you believe and how
you calculate it) panicking about write management on an SSD is really
over the top isn't it? Seriously we are looking at read/write
lifetimes equal or exceeding what we see on magnetic media...

Or am I totally misinterpreting and misunderstanding here?

C.
--
To unsubscribe, e-mail: opensuse+unsubscribe@xxxxxxxxxxxx
To contact the owner, e-mail: opensuse+owner@xxxxxxxxxxxx



Relevant Pages

  • Re: SSD: Anyone know if this applies to other manufacturers
    ... In the early days of flash ram (cards and pen drives) without ... as high as 100,000 cycles. ... The SLC nand chips used by SSDs with wear levelling controllers, ... This might seem to be a worrying trend regarding SSD 'lifetimes' but ...
    (uk.comp.homebuilt)
  • Re: [opensuse] SSD in openSUSE.
    ... still need this careful management of write cycles? ... with current SSDs and their write cycle ratings (a current/new SSD ... you are looking at 5,000,000 or more write cycles now), ...
    (SuSE)
  • Re: To homebuild or not to homebuild 2
    ... It's a fortunate fact that SSDs fail on write cycles, ... That rather depends on whether the blocks on a flash chip should all ... The memory controllers can only spread wear over the unallocated ... Occasionally erasing all the files on an SSD and starting ...
    (uk.comp.homebuilt)
  • Re: To homebuild or not to homebuild 2
    ... It's a fortunate fact that SSDs fail on write cycles, ... That rather depends on whether the blocks on a flash chip should all ... Occasionally erasing all the files on an SSD and starting ... At least Defrag shouldn't be necessary so that's less wear. ...
    (uk.comp.homebuilt)