Berkey B Walker wrote:
Mikael Abrahamsson wrote:
On Wed, 21 Apr 2010, Bill Davidsen wrote:
I hear this said, but I don't have any data to back it up. Drive
vendors aren't stupid, so if the parking feature is likely to cause
premature failures under warranty, I would expect that the feature
would not be there, or that the drive would be made more robust.
Maybe I have too much faith in greed as a design goal, but I have to
wonder if load cycles are as destructive as seems to be the assumption.
What I think people are worried about is that a drive might have X
load/unload cycles in the data sheet (300k or 600k seem to be normal
figures) and reaching this in 1-2 years of "normal" (according to the
user who is running it 24/7) might be worrying (and understandably so).
Otoh these drives seem to be designed for desktop 8 hour per day use,
so running them as a 24/7 fileserver under linux is not what they
were designed for. I have no idea what will happen when the
load/unload cycles goes over the data sheet number, but my guess is
that it was put there for a reason.
I'd love to find some real data, anecdotal stories about older
drives are not overly helpful. Clearly there is a trade-off between
energy saving, response, and durability, I just don't have any data
from a large population of new (green) drives.
My personal experience from the WD20EADS drives is that around 40% of
them failed within the first year of operation. This is not from a
large population of drives though and wasn't due to load/unload
cycles. I had no problem getting them replaced under warranty, but
I'm running RAID6 nowadays :P
Sorry, you sound like a factory droid. *I* see no reason for early
failure besides cheap mat'ls in construction. Were these assertations
of short life to be true, I would campaign against the drive maker.
(I think that they are just normalizing failure rate against warranty
claims) Buy good stuff. I *wish* I could define the term by mfg. It
seems Seagate, & WD don't hack it. The Japanese drives did, but since
the $ dropped -
Let's see, first you put my name on something I was quoting (with
attribution), delete the correct name of the person you are quoting, and
then call me a "factory droid." So I have some idea of your attention to
detail. Second, the short term failure rates are influenced by
components delivered, assembly, and treatment in shipping. So assembly
is controlled by the vendor, parts are influenced by supplier selected,
and delivery treatment is usually selected by the retailer. A local
clone maker found that delicate parts delivered on Wednesday had a
higher infant mortality that other days. Regular driver had Wednesday
off, sub thought "drop ship" was an unloading method, perhaps.
One thing seemingly missed is the relationship between storage density
and drive temp.variations.. Hard drive mfgs are going to be in deep
doodoo when the SSD folks get price/perf in the lead lane. This year,
I predict. And maybe another 2 for long term reliability to be in the
lead..
I think you're an optimist on cost equality, people are changing to
green drives which are generally slower due to spin down or lower rpm,
because the cost of power and cooling is important. It's not clear if
current SSD tech will be around in five years, because there are new
technologies coming which are inherently far more stable for multiple
writes. The spinning platter may be ending, but the replacement is not
in sight. In ten years I doubt current SSD tech will be in use, replaced
by phase change, optical isomers, electron spin, or something still in a
lab. And the deployment of user visible large sectors (write chunks,
whatever) is not clear, if the next tech will work just as well with
smaller sectors, this may become a moot point.
I believe that many [most?] RAID users are looking for results (long
term archival) that are not intended in the design.We are about 2
generations away from that being a reality - I think.. For other
users, I would suggest a mirror machine. with both machines being
scrubbed daily, and media being dissimilar in mfg and mfg date.
It's not clear that date of manufacture is particularly critical, while
date of deployment (in-use hours), probably is. But looking at the
Google disk paper, a crate of drives from the same batch doesn't all
drop dead at once, or close to it, so age in service is a factor, but
not likely a critical factor.
I can't wait until Neil gets to (has to) play/work with the coming
tech. Neat things are coming.
I would rather see some of the many things on the "someday list" get
implemented. It's more fun to play with new stuff than polish off the
uglies in the old, but the uglies are still there.
--
Bill Davidsen <davidsen@xxxxxxx>
"We can't solve today's problems by using the same thinking we
used in creating them." - Einstein
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
