| 1 |
On Sunday 28 September 2008 00:18:35 Stroller wrote: |
| 2 |
> > Since they are not frequently updated and have minimal installed |
| 3 |
> > software |
| 4 |
> > (iptables on firewalls and DNS on DNS servers) accompanied by the |
| 5 |
> > fact that most devices have internal wear leveling; it should take |
| 6 |
> > many years to reach the write cycle limits? |
| 7 |
> |
| 8 |
> I've read a fair little bit about this subject and never gotten a |
| 9 |
> definitive answer on what is "safe", but AIUI the wear-levelling on |
| 10 |
> flash memory is filesystem-dependent. Thus it may work fabulously well |
| 11 |
> for FAT filesystems, and not at all for EXT. |
| 12 |
|
| 13 |
Rule of thumb: |
| 14 |
|
| 15 |
The problem is that the ability for individual memory cells to reliably |
| 16 |
perform writes deteriorates over time. Cheap and nasty devices can start to |
| 17 |
fail after 10,000 writes to a cell, the better devices can often cope with |
| 18 |
100,000 writes to a single cell. |
| 19 |
|
| 20 |
The reason there is little definitive data is that it isn't a definitive |
| 21 |
problem - the variables vary wildly. Like you say, some filesystems do wear |
| 22 |
levelling (some better than others), some use cases are frugal with their |
| 23 |
writes, and the device itself has enormous variance as to when it will stop |
| 24 |
performing as expected. |
| 25 |
|
| 26 |
The numbers above must be interpreted as the maximum number of writes where |
| 27 |
the manufacturer is still prepared to guarantee the device. |
| 28 |
|
| 29 |
-- |
| 30 |
alan dot mckinnon at gmail dot com |
Archives