RSYNC Fedora (Moving onto RAID...)
ART KAGEL, BLOOMBERG/ 65E 55TH
KAGEL at bloomberg.net
Fri Nov 28 15:47:31 UTC 2003
Scott Burns said:
> William Hooper wrote:
>>
>>>Of course if both disks of a mirror set die you lose everything, too.
>
>> Yes.
>
>> The guy was recommending RAID10 ( a bunch of RAID1 pairs then add them
>> all together as RAID0 ).
'The guy' is me.
>> After reading it I worked out a number comparing a RAID5 with 5 disks
>> against RAID10 with 4 stripped sets of 3 disks (12 in total ). If you
>> lose 2/5 RAID5 disks you are all lost.
I only use 2-way mirrors in my RAID10 arrays, but a 3-way mirror just increases
the safety advantages of RAID10 over RAID5, though at a greatly increased cost.
> Apples and oranges. If you are going to compare multiple sets, why not
> compare 4 RAID5 sets w/ 3 disks each (12 in total).
Huh? Now you're comparing apples to sausages. RAID5 purports to provide data
redundancy with a minimum of disks. A single 5-drive RAID5 array provides 4
drives worth of data storage for the cost of a 5th drive of parity (yes parity
is distributed, but the storage penalty/cost of the array is one drive
additional to provide space for parity/redundance). Compare that to a single 4
pair RAID10 array which costs you 4 drives for redundancy to obtain the same
storage capacity. That's not apples to oranges, it's macintosh apples -versus-
granny smith apples. Like that argument we can compare the flavor, texture,
sweetness, quantity of juice produced, etc. without getting bogged down in
whether it is a fair comparison because the one is red and the other green.
> If you lose 5/12 RAID10s you
> still have over a 95% chance of no data lost. I believe recovery time
> on RAID10 puts RAID5 to shame too. His entire rant can be found at
But we're not talking about losing five drives, We're talking about losing two
drives out of 5 versus 2 drives out of 8. Yes the probability of losing the
second drive out of the remaining 4 RAID5 drives is about 40% less likely than
losing the second drive out of the remaining 7 RAID10 drives. However, the
difference is that in a RAID10 you only suffer data loss if a failed drive's
particular mirror fails before it can be replaced and recovered (and in the
3-way mirror example BOTH of the particular mirror drives). Compare this to the
RAID5 case where the loss of ANY of the remaining 4 drives will cause
catastrophic and complete data loss. This makes the probability of dataloss
MUCH higher in the RAID5 array than in RAID10.
<SNIP>
> He answers your question there:
> "Well with RAID10 there is no danger unless the one mirror that is
> recovering also fails and that's 80% or more less likely than that any
> other drive in a RAID5 array will fail!"
Of course, he falls into the same trap you did, comparing a single RAID5
set with multiple RAID10 sets.
NO! I actually do use multiple RAID10 arrays striped again to create what is
known as a 'Plaid', but we're not discussing that technique which is only used
to provide MASSIVE storage and MASSIVE throughput in a single logical drive.
We're discussing recovery and data redundancy here, so we're comparing a single
RAID5 array to a single RAID10 array.
Perhaps you are not familiar with exactly what RAID10 is? A RAID10 array is a
strip (RAID0) formed from a number of mirrored pairs (or triplets) (RAID1).
Since the RAID1 is applied first over which RAID0 is applied it is known as
RAID10. The opposite, mirrored stripe sets, is known as RAID01.
This discussion of data loss due to second drive failure is a minor part of the
problems I have with RAID5 which include 50% write performance penalty and
absolutely NOT PROTECTION at all against partial media failure causing partial
data loss.
BTW in one of the postings someone said something like "and besides, we all do
take backups and that's what backups are for" when discussioning catastrophic
array failure. This is shortsighted. Do we all have sufficient disks in
inventory to restore and recover from backups in a manner timely enough to
prevent lost business? Can we afford the down time? This does not even address
the loss of transactions to the files/databases that have not yet been
archived! The loss of transactions can be catastrophic to the integrity of the
organization and its relationship to customers! If we are discussing a bank,
can the bank say to you: "Oh, gee, we're sorry your deposit did not show up in
your account, Our disks crashed yesterday and your transaction wasn't on the
backup tapes we restored. Just make the deposit again!" Would you remain a
customer at that bank? Or to be less rediculous, could they say: Sorry you
cannot make a withdrawal at this time. Please come back in 5 hours when our
system drives have been restored from backup." Or: "Please come back in 3 days
when we've received our rush order of new disks. Not to worry everything is
safely on tape!" No!
Anyone interested in the original postings, I have reposted my enhanced
Quarterly RAID5 Rant to comp.databases.informix (which group - CDI - is archived
on the International Informix Users' Group site www.iiug.org). Anyone
interested in the evils of RAID5 can checkout several articles on the subject by
myself and other authors at www.baarf.com (BAARF is a Denmark based group
fighting this good fight).
Art S. Kagel
More information about the users
mailing list