Current disk failure stats from Russia

  • Thread starter Thread starter Arno
  • Start date Start date
A

Arno

Ha, found it again. Pretty interesting and conclusive. They do
not fear naming names, I guess they are not afraid of being sued.

They do have failure rates per manufacturer in relation to sales
and also have insights into age at failure per manufacturer.

While the sample of 4000 drives is not large enough to qualify
as scientifically solid, it is IMO the best available data
at the moment and (again) shows that the relevant publication
by Google was wrong not to separate their numbers by manufacturers
(among other things that make the Google results pretty doubtful).
In addition, these drives were operated in a variety of different
conditions, which also helps relevancy.

Executive summary:
Seagate: stay away, WD: so-so, Hitachi: best by a fair margin

Here is the (english) story on tomshardware:

http://www.tomshardware.com/reviews/hdd-reliability-storelab,2681.html

Here is the Russian link:

http://www.storelab.ru/sravnenie-nadezhnosti-hdd.htm


Arno
 
Timothy said:
And, of course, I'll save on RAM by using a SATA 3 SSD -
probably made by Crucial - for the swap file.
Click to expand...

Do you really think that's a good idea? Or were you joking?
 
Timothy said:
Using Crucial is a good idea. :-)
Click to expand...

It is, I've always bought Crucial (in fact, I have my eye on one of
their SSDs too, fast read speeds for loading OS and apps.)

But using an SSD for swap isn't a good idea.
 
Mike Tomlinson said:
It is, I've always bought Crucial (in fact, I have my eye on one of
their SSDs too, fast read speeds for loading OS and apps.)
Click to expand...
But using an SSD for swap isn't a good idea.
Click to expand...


Historically, you are quite right. But with modern wear-leveling
it starts to be a better idea, if the swap area is significantly
smaller than the disk.

Arno
Click to expand...
 
But not statistical accuracy, since what each one died of
should become a separate category of failure mode and a
smaller statistical universe in which to judge each failure rate.
Click to expand...

I agree. And you would need to look at the full popultaion, how each
drive was handled and what load it was under.
(But, as you say, this is probably the best data we can get.)
Definitely.

I've used only Maxtor (now owned by Seagate) HDs in the
past with no failures at all (that I know of), but I think I'll go
Click to expand...

I had about 50 Maxtors (the problematic ones) in a server cluster
I built, whith no failures whatsoever except for a few drives
inadequately packaged and dropped in shipping. However these
were well cooled and surface-scanned every 14 days.

I think Maxtors are just not resilient to abuse drives suffer
with ordinary consumers, especially wtith regard to inadequate
cooling. That their own external drices are inadequately cooled
makes this worse. They are perfectly fine in a server-room
environment.
with Hitachi for rotaional HDs in the future based on this data.
Click to expand...

I have a lot of WDs now, I like their cheap external drives, were
you get the disk and a quite reasonable enclosure for the price
of the bare drive.
And, of course, I'll save on RAM by using a SATA 3 SSD -
probably made by Crucial - for the swap file. <hee, hee>
Click to expand...

;-)=)

Arno
 
Arno <[email protected]> said:
Historically, you are quite right. But with modern wear-leveling
it starts to be a better idea, if the swap area is significantly
smaller than the disk.
Click to expand...

Agreed, though I think if the machine is starting to swap in normal use
it's better to add more memory than to swap to SSD. (Yes, Tim, I know
you were saying that).

I personally would always put the swap on a spinning disk along with
data and keep the speed of the SSD for loading the OS and apps.
 
Mike Tomlinson said:
Agreed, though I think if the machine is starting to swap in normal use
it's better to add more memory than to swap to SSD. (Yes, Tim, I know
you were saying that).
Click to expand...

Definitely! I see the main use for swap now memopry pages that
get allocated but hardly ever used, e.g. in demon processes.
That is alsow hy the rule "swap size = memory size" is pretty
outdated. I find that 100...256MB are quite enough for modern
OSes (sorry, MS does not qualify, not even Win7).
I personally would always put the swap on a spinning disk
along with data and keep the speed of the SSD for loading
the OS and apps.
Click to expand...

There is a case where it makes sense to put swap on an SSD,
namely where you have no choice, e.g. with a hard memory
linit and you are forced to user more, distributed over
several processes. That is pretty rare though. In doubt,
go for more RAM.

Arno
Click to expand...
 
Timothy Daniels said:
"Mike Tomlinson" advised:

Hey! Using a small SSD for a swap (paging) file sounds cool -
which should be reason enough to use it! I mean, more RAM is
good, but it's just more or the same thing - nothing to talk about.
Using a 50GB SSD for swapping parts of those REALLY LARGE
files is both occasional RAM backup for those monumental editing
jobs and a cool feature to impress clients with. Just say it's part of
your Graduated Response Storage (GRS) System, and the contract
will be yours. :-)
Click to expand...

Arggggghhhhhh! ;-)
And, of course, if your system architecture limits words to 32 bits
and addressing space to 4GB, more RAM than that isn't an option.
Click to expand...

Indeed. I see this as the only really legitmate use.

Arno
 
Timothy said:
And, of course, if your system architecture limits words to 32 bits
and addressing space to 4GB, more RAM than that isn't an option.
Click to expand...

A very good point, hadn't thought of that. Thanks.
 
Timothy said:
"Mike Tomlinson" advised:


Hey! Using a small SSD for a swap (paging) file sounds cool -
which should be reason enough to use it! I mean, more RAM is
good, but it's just more or the same thing - nothing to talk about.
Using a 50GB SSD for swapping parts of those REALLY LARGE
files is both occasional RAM backup for those monumental editing
jobs and a cool feature to impress clients with. Just say it's part
of your Graduated Response Storage (GRS) System, and the contract
will be yours. :-)

And, of course, if your system architecture limits words to 32 bits
and addressing space to 4GB, more RAM than that isn't an option.
Click to expand...

Yes it is, for a virtual drive just for the swap file, much faster than an SSD and no wear possibility either.
 
Mike Tomlinson said:
A very good point, hadn't thought of that. Thanks.
Click to expand...
PAE allows 36 (IIRC) *physical* address bits in page table entries.
Don't know about Windows, but Linux can use the additional highmem.

Jerry
 
Ha, found it again. Pretty interesting and conclusive. They do
not fear naming names, I guess they are not afraid of being sued.
Click to expand...

The Russkie talks big, but you can't expect a bunch of ignorant
peopns to understand a machine the way our boys can.
 
Back
Top