Write cycles are just as damaging as erase cycles. And, it is certain
disreputable manufacturers of storage products, basically only packagers, who
are claiming 100k cycles. The chip manufacturers generally claim 10k. Now with
wear levelling it may seem that they will last quite a long time, but it depends
entirely on how much free space there is on the device. If, in the extreme case,
only one sector remains free, 10k writes will kill it. I have seen some other
incorrect calculations not so long ago, which also made the same error of using
total device capacity in the calculation. It is not, it is always the free space
that matters, for any solid state memory with wear levelling. Nevertheless, if
care is taken to ensure that there is always a large amount of free space, say
50%, and the kernel "swappiness" is set to minimum, solid state storage devices,
such as USB sticks or SD cards, are fit for the envisaged purpose, where a life
of a few years is all that will be required. But ferroelectric RAM (FRAM), which
is non-volatile and has a much larger capability for write cycles (also per
byte, not per sector) is coming along and when its storage density becomes
comparable to flash, the limited life will largely disappear. I am not convinced
that anyone knows yet just how many write cycles a FRAM cell will tolerate, and
whether it is infinite or just a very large number, but it is certainly much
better than flash. And, it works more or less like RAM, i.e. almost as fast to
write as to read, and every bit independently changeable. See here for example. [ Reply to This | Parent | # ]
|