I am planning to build a RAID array for long life file server.  With
upgrades in the operating systems and computer hardware becoming more
and more rapid, I would like to be able to host all the home
directories on this RAID server so that the client machines can be
changed or upgraded as needed.

I have done a lot of research all over and for the most part I have
found the answers.  One question bothers me still.

I had read that in order to prevent failure of multiple disk drives in
an array at the same time, one should purchase drives from different
manufacturing batches.  Can anybody suggest how to go about it ?  I
mean, a company like Seagate sells about forty million disk drives
every year, so, how long should I wait between my disk drive purchases
?

Thank You,

Devang.

[ Post a follow-up to this message ]

On 26 Oct 2006 11:45:23 -0700, "Big D" <devangdevang@netscape.net>
wrote:

>I am planning to build a RAID array for long life file server.  With
>upgrades in the operating systems and computer hardware becoming more
>and more rapid, I would like to be able to host all the home
>directories on this RAID server so that the client machines can be
>changed or upgraded as needed.
>
>I have done a lot of research all over and for the most part I have
>found the answers.  One question bothers me still.
>
>I had read that in order to prevent failure of multiple disk drives in
>an array at the same time, one should purchase drives from different
>manufacturing batches.  Can anybody suggest how to go about it ?  I
>mean, a company like Seagate sells about forty million disk drives
>every year, so, how long should I wait between my disk drive purchases
>?
>
>Thank You,
>
>Devang.

Losing multiple drives in an array at the same time is not really the
issue, it's losing multiple drives in the same raid group that will
kill you.  Avoiding that is what every storage admin hopes to do, but
I'm not aware of any mathematical algorithm that will tell you how to
do it.

Keep the raid groups small and use raid_dp or raid 6 or raid10.  All
of those will minimize the risk of double disk failure tremendously.
Of all of them raid 10 is the safest, and most costly as logic would
have it.  Avoid raid 01 as that will kill you far faster than anything
(except plain 'ol raid 0 of course), though I believe no respectable
vendor uses that anymore.

Make sure to have a good ratio of spares to data drives.  I use a
ratio of 28:1 with raid_dp.  No issues so far.

~F

[ Post a follow-up to this message ]

Faeandar wrote:

...

> Keep the raid groups small and use raid_dp or raid 6 or raid10.  All
> of those will minimize the risk of double disk failure tremendously.
> Of all of them raid 10 is the safest, and most costly as logic would
> have it.

I think your fingers got a bit ahead of your brain above.  RAID-DP and
RAID-6 are *far* 'safer' than RAID-10 in surviving multiple disk
failures (or a single disk failure after which you discover that you've
got a bad sector somewhere else):  while RAID-10 can *potentially*
survive the failure of half the disks, unless your array is truly
gargantuan the likelihood that a second drive (or sector) failure will
be the partner of the first drive that failed (and thus you'll lose data
or have to restore from backup) is far higher than the likelihood that
*3* drives will fail (or experience critical bad sectors) in a RAID-DP
or RAID-6 configuration (which is what it takes there to lose data or
have to restore from backup).

RAID-6 and RAID-DP have good enough availability characteristics that
'keeping raid groups small' is not much of an issue, unless (as I noted
above) they grow to gargantuan sizes (without going through the numbers,
I suspect that dozens of disks in a single RAID-6 or RAID-DP group would
be just fine, maybe even hundreds).  The main difference between these
options and RAID-10 is performance for smallish updates, where parity
RAIDs tend to be significantly slower (though large, non-volatile
controller caches can help with that problem).

- bill

[ Post a follow-up to this message ]

In answer to Big D's question, I don't know how you go about buying
disks from different processing batches and frankly I think even
worrying about it is overkill for a home network. I mean enterprise
storage systems are filled with hundreds of disks presumably from the
same processing batch and its no big deal.

I think as long as you implement RAID 5 or 6 depending on how paranoid
you are about a disk failure then you will be fine. As long as you keep
ontop of any failed disks. Get your system to email or something if it
detects a failed disk.

I have a linux server in my garage with over 1TB of disk capacity
RAIDed together with software RAID 5 and has been serving files to 5
pcs in my house for a few years now with no problems at all. All the
disks were purchased from the same place at the same time.

[ Post a follow-up to this message ]

Stunster wrote:
> In answer to Big D's question, I don't know how you go about buying
> disks from different processing batches and frankly I think even
> worrying about it is overkill for a home network. I mean enterprise
> storage systems are filled with hundreds of disks presumably from the
> same processing batch and its no big deal.

Our expensive system (NetApp) has disks from multiple vendors and
revisions. Our cheaper systems (Adaptec) have all identical drives (and
they fail like crazy, and the vendor complained because we 'overuse' the
FC drives..??).

> I have a linux server in my garage with over 1TB of disk capacity
> RAIDed together with software RAID 5 and has been serving files to 5
> pcs in my house for a few years now with no problems at all. All the
> disks were purchased from the same place at the same time.

You are with the 90 % of lucky guys. If you are with the 10% of unlucky
users to have 90% of the failures... 



Thomas

[ Post a follow-up to this message ]

On Thu, 26 Oct 2006 23:17:34 -0400, Bill Todd <billtodd@metrocast.net>
wrote:

>Faeandar wrote:
>
>...
> 
>
>I think your fingers got a bit ahead of your brain above.  RAID-DP and
>RAID-6 are *far* 'safer' than RAID-10 in surviving multiple disk
>failures (or a single disk failure after which you discover that you've
>got a bad sector somewhere else):  while RAID-10 can *potentially*
>survive the failure of half the disks, unless your array is truly
>gargantuan the likelihood that a second drive (or sector) failure will
>be the partner of the first drive that failed (and thus you'll lose data
>or have to restore from backup) is far higher than the likelihood that
>*3* drives will fail (or experience critical bad sectors) in a RAID-DP
>or RAID-6 configuration (which is what it takes there to lose data or
>have to restore from backup).
>
>RAID-6 and RAID-DP have good enough availability characteristics that
>'keeping raid groups small' is not much of an issue, unless (as I noted
>above) they grow to gargantuan sizes (without going through the numbers,
>I suspect that dozens of disks in a single RAID-6 or RAID-DP group would
>be just fine, maybe even hundreds).  The main difference between these
>options and RAID-10 is performance for smallish updates, where parity
>RAIDs tend to be significantly slower (though large, non-volatile
>controller caches can help with that problem).
>
>- bill


My fingers did not get ahead of my brain, but my general environment
clouded my view.  My disk environment is fairly large so generally I
think in terms of 160+ drives at a time.  Hence raid 10 is usually
better.  However, it is not an option for me so my experience with it
is actually from days of yore where the environment was smaller, but
raid_dp and raid6 were not yet avaialble.  Worlds colliding and
such...

For the OP, Bill is correct.  Raid_dp or raid 6 (depends on which
vendor you talk to) are very safe, even compared to raid 10.  I doubt
anyone would have a 100+ raid group, regardless of raid type, since 1)
there's no real performance benefit after 50 or so drives and B) the
failure liklihood is just too high still.  Some day I'm sure it will
be commonplace, but not today.

~F

[ Post a follow-up to this message ]

Big D wrote:
> I am planning to build a RAID array for long life file server.  ...
> I had read that in order to prevent failure of multiple disk drives in
> an array at the same time,

If you are seriously concerned about coincident lifespans then there is
a simple and effective preventative: Rotate out and replace some of
your drives (e.g. half mirrors) after e.g. 6 or 12 months. Your swapped
out set, if complete, can make up a functional emergency backup.

> one should purchase drives from different
> manufacturing batches.  Can anybody suggest how to go about it ?  I
> mean, a company like Seagate sells about forty million disk drives
> every year, so, how long should I wait between my disk drive purchases
> ?
>
> Thank You,
>
> Devang.

[ Post a follow-up to this message ]

I think I have got many great answers here, and I thank you all.

Toby in particular had a very useful suggestion, rotate a drive every
six or twelve months and use the old one as a backup.  It works for me
since I wouldn't mind creating two arrays over time and use the second
array for my home media server, that way I can separate out work and
play.

The rest of you have also provided great suggestions, so the thanks go
out for all of you.

Take Care,

Big D

[ Post a follow-up to this message ]