[OpenIndiana-discuss] OpenIndiana-discuss Digest, Vol 33, Issue 27

[Sebastian Gabler]

Am 17.04.2013 11:16, schrieb "Edward Ned Harvey (openindiana)" 
<openindiana at nedharvey.com>
> It's a fact that NAND has a finite number of write cycles, and it gets slower to write, the more times it's been re-written.

AFAIC, these are two facts, and the latter is much more relevant in 
production. Someone mentioned it earlier: You need SSDs that have a 
predictably, stable, low write latency.
However, in that use case the majority of writes are bulk, not 
transactional, as far as I have understood. So, even if the disk 
controller will slow down writes, the worst thing that should is that he 
has to wait longer for the bulk to finish.

For the life span of NAND, the typical behaviour is that a certain group 
of blocks cannot be written to. It is not common that resident data 
becomes irretrievable. For a non-junk drive, we should assume that it 
can handle all that transparently to ZFS.

BR

Sebastian