[lug] USB "wear out" was: dd ssh pipe to bzip?

Michael J. Hammel mjhammel at graphics-muse.org
Thu Nov 18 08:52:41 MST 2010


On Thu, 2010-11-18 at 05:28 -0700, Stephen Queen (actually someone else)
wrote:
> > Is it an urban myth that flash media will "wear out"?  

Not a myth.  MTD devices have limited writes, though these range up to
100000 or more these days.  In the early days of MTD the writes could be
limited to 1000-10000.

The problem is that MTD doesn't write a single memory location.  When an
MTD is written it has to write a larger chunk of data (I don't remember
off hand the size - I think its a block, based on the MTD
configuration).  So, for example, updating an /etc/hosts
or /etc/resolv.conf with a new entry when your DHCP changes can be very
rough on a USB stick or SD card.  That's why some distros remap certain
areas, such as /var, to a tmpfs at boot time.  This was one of the first
things I did with BeagleBox so I could extend the life of the SD card
that I'm using for testing.

> I've seen things that indicate those cycles are approximately
> > the same as the number of cycles you get on a hard drive platter. 

SSD devices may be pushing this limit.  Large data centers are switching
to SSDs for speed and throughput improvements.  But I've not heard if
they've seriously increased the write cycles.

>  In
> > other words, there's no longer any real difference between writes on
> > USB and HDD.

A USB stick is not in the same class as the larger scale SSDs.  It still
has limited write cycles.  It's a matter of cost.  The parts for the
limited life cycle flash is much lower than the SSD, which can be very
expensive.  A USB stick probably won't survive as many write cycles as a
hard disk.

> > In addition, the number of writes is some huge number per cell and
> it's
> > unlikely you'd ever hit that unless you specifically had a program
> > constantly writing to exactly the same location on the drive...

Not with MTD devices because of the way a write has to be a block or
larger at a time.  You don't have to actually be updating a cell to
cause a write to it.

> >This scrolled across reddit just the other day:
> 
> >http://www.bress.net/blog/archives/114-How-Long-Does-a-Flash-Drive-Last.html

That looks about right, based on what I've been reading elsewhere: 30k
writes till failure.  Under 100k.  Yeah, sounds about right.

-- 
Michael J. Hammel                                    Principal Software Engineer
mjhammel at graphics-muse.org                           http://graphics-muse.org
------------------------------------------------------------------------------
When C++ is your hammer, everything looks like a thumb.
              -- Steven M. Haflich




More information about the LUG mailing list