On 200802231201, Michelle Konzack wrote:
> Am 2008-02-20 20:33:12, schrieb Anders Breindahl:
> > If energy consumption of hard drives is important, consider flash-based
> > storage, which are an order of magnitude more economical. (But take a
> > stand with regards to flash outwearing).
>
[snip]
> > Actually---how mature do you, the debian-isp folks, consider flash-based
> > storage for server usage? Obviously, one would shred the flash cells for
>
> Storage Server on CF? -- It's a joke!
> Maybe it works for VERY static data but not for a ${HOME} directory...
Why not, then?
For example, Lenovo's new X300 laptop is equipped solely with a SSD for
nonvolatile storage. Apparently they consider SSD's to be mature enough
for $HOME, as you put it.
Moreover, there's no wear in *accessing* flash memory, so for server
applications where content assumably will be read way more often that
it's written, I see no particular wear. I do however identify wear in
sqld's and smptd's, for example.
As stated on
http://en.wikipedia.org/wiki/Solid-state_drive#Disadvantages
one can attempt to mitigate the problem of wear in areas such as log
files by---even at firmware level---do wear levelling. I don't know to
what extent this can be done, and to which extent it actually is done.
However,
http://wiki.eeeuser.com/ssd_write_limit
suggests that 50% wear levelling is below industry standard. (But is it
sacrificing capacity do do so?---can't tell.)
Also, that link shows some interesting wear statistics. I'm reading that
with below-industry-standard wear levelling of 50%, one could
continuously shred the drive of an Eee pc for 2 to 13 years before cells
would fail. And that's for 10^5 write cycle drives. The Wikipedia
article mentions that 10^6 up to 5*10^6 write cycle drives exist.
I.e., other factors omitted, you could continuously push data onto such
a 4GB 10^6 high-endurance drive at 50% levelling at 10MB/s for
4*10^9/10^7*10^6 * 0.5 / (3600*24*365.2524) = 6.3 years
Or taking advantage of the capacity for wear levelling, a 64GB of Eee PC
quality (matching the X300's, assumably):
64*10^9/10^7*10^5 * 0.5 / (3600*24*365.2524) = 10.1 years
Why would this not suffice for server applications? Especially
considering the good random access times of SSDs, and the fact that we
will use these drives in RAIDs anyway?
Regards, skrewz.
Attachment:
signature.asc
Description: Digital signature