[Gllug] RAID 1+0 vs 0+1

[Steve Nelson]

Hello Chums,

A debate arose on irc as to whether there's a difference between RAID
1+0 and 0+1.  I contend that 1+0 is more resilient.  Our esteemed Mr
Brooks disagreed, and stated "There is no difference".  As my attemtps
to draw ASCII depictions of disks kept getting interrupted, I  thought
I'd email to the list instead.

Let's take a machine with 6 x 9G disks in it.  I'm arguing that if we
mirror first,  then stripe, we have a more resilient setup.  In my
diagrams, letters represent mirrors and numbers disks.

Option 1:  Stripe, then Mirror - 0+1

A0 <-> A1 <-> A2
B0 <-> B1 <-> B2

In this case, we have a stripe of mirrors.  If we lose any disk in
either stripe, we kill the entire stripe.  We're now left with only
one striped array.  We can now onl
y tolerate one more disk failure on the other stripe.

Option 2:  Mirror, then Stripe - 1+0

A0 B0 C0
 | |  |
A1 B1 C1

In this case, if we lose any disk, say A0, we can afford to lose two
more disks before we're in trouble.  We could lose, for example,
either A0 or A1, then any disk from B and any disk from C.  This is
only slightly more resilient, and we can't predict the order in which
the disks will fail, but it is different, and it is more resilient.
The easy way to remember this is with the little saying: "Remember,
M&S is safer than S&M!"

Would appreciate comments / criticisms.

Thanks,

S.Hello Chums,

A debate arose on irc as to whether there's a difference between RAID
1+0 and 0+1.  I contend that 1+0 is more resilient.  Our esteemed Mr
Brooks disagreed, and stated "There is no difference".  As my attemtps
to draw ASCII depictions of disks kept getting interrupted, I  thought
I'd email to the list instead.

Let's take a machine with 6 x 9G disks in it.  I'm arguing that if we
mirror first,  then stripe, we have a more resilient setup.  In my
diagrams, letters represent mirrors and numbers disks.

Option 1:  Stripe, then Mirror - 0+1

A0 <-> A1 <-> A2
B0 <-> B1 <-> B2

In this case, we have a stripe of mirrors.  If we lose any disk in
either stripe, we kill the entire stripe.  We're now left with only
one striped array.  We can now onl
y tolerate one more disk failure on the other stripe.

Option 2:  Mirror, then Stripe - 1+0

A0 B0 C0
 | |  |
A1 B1 C1

In this case, if we lose any disk, say A0, we can afford to lose two
more disks before we're in trouble.  We could lose, for example,
either A0 or A1, then any disk from B and any disk from C.  This is
only slightly more resilient, and we can't predict the order in which
the disks will fail, but it is different, and it is more resilient.
The easy way to remember this is with the little saying: "Remember,
M&S is safer than S&M!"

Would appreciate comments / criticisms.

Thanks,

S.
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug