[ale] no putting swap on ssd

Brian Pitts brian at polibyte.com
Mon Feb 7 21:39:57 EST 2011


On 02/07/2011 05:20 PM, Greg Freemyer wrote:
> On Mon, Feb 7, 2011 at 4:52 PM, Sparr <sparr0 at gmail.com> wrote:
>> This is mostly bogus out-of-date information. Yes, SSDs (and other
>> solid state storage like SD cards) have a limited lifetime, but it far
>> exceeds their useful lifetime these days.
>>
>> Assume we have wear leveling, so that single sectors don't go bad prematurely.
>>
>> At 50 megabytes per second of continuous writing (which is insane for
>> any storage device except always-on video recording), a 16GB storage
>> device with a lifetime of 1 million writes means it's going to fail in
>> TEN YEARS.
> 
> I guess I'm way behind the times thinking its 10,000 writes per EB.  I
> just pulled up the Intel® X25-V SSD spec. to see what it is, but it
> doesn't have the number of writes provided in the spec. :(
> 
> I'll assume your 1,000,000 number is right even in the "value" SSDs.
> 
> But for clarity, most storage apps have a lot of static data involved.
> 
> So let's assume the first thing I do is put 12GB of static data on
> that 16GB device.  And then never write to that data again.  That
> removes that 12GB from the wear-leveling equation.
> 
> So now I only have 4GB (plus some small amount of spare EBs) with
> which to wear-level.
> 
> So instead of 10 years, its 2.5 years.  Next if your partition is not
> properly aligned it can easily cause 2 real writes for every userspace
> initiated write.    So now it's 1 1/4 years.  (New parted as an
> example recommends partitions start on 1 MiB boundaries.  That should
> properly align.  Parted from before Dec. 2009 tried to force alignment
> on cylinder boundaries.  That would often be misaligned for SSDs.)
> 
> I'm not seriously saying you'll see SSDs simply wear out after a year
> or so, but I am saying it's something to be concerned about and plan
> around or you may inadvertently find yourself behind the 8-ball.

If you're interested in but unfamiliar with this topic, I recommend reading

http://www.anandtech.com/show/2829/1

Some quotes relevant to Greg's post:

"Each MLC NAND cell can be erased ~10,000 times before it stops reliably
holding charge. You can switch to SLC flash and up that figure to
100,000, but your cost just went up 2x. For these drives to succeed in
the consumer space and do it quickly, it must be using MLC flash."

"But remember, today’s consumer drives only ship with roughly 6 - 7%
spare area on them... By comparison, the enterprise SSDs like Intel’s
X25-E ship with more spare area - in this case 20%."

-- 
All the best,
Brian Pitts


More information about the Ale mailing list