[geeks] RAID quirks and terminology

[Phil Stracchino]

On 06/15/14 12:29, Lionel Peterson wrote:
> I'm confused.
> 
> A2 was a mirror of A1, or part of a stripe pair with A1? How does striping
> figure into this?

Three mirrors:  A1/C2, A2/B1, C1/B2.  The mirrors then get striped
together (instead of just a linear concat) to distribute writes as
evenly as possible across the devices.

> If you wrote a file onto this array, how many copies would there be?

Two.

> Assuming each drive has the same capacity ('n'), what was the total storage
> available to the user? I'm guessing 3n/2? RAID5 would give you 2n wouldn't
> it?

3n/2, correct.  360MB from three 240MB drives, iirc.

> Why did you think this was better than RAID5? Lower computational overhead?

Yup.  RAID5 on that hardware would have been prohibitively slow.  This
hack got me a 50% increase in storage with very little I/O performance
penalty.

> Why wouldn't this be RAID5? This is how I describe RAID5 to folks...

Because it's pure mirroring and striping with no parity calculation
involved.



>> On Jun 15, 2014, at 10:54 AM, Phil Stracchino <phils at caerllewys.net> wrote:
>>
>> I did a rather unconventional mirror-and-stripe once that wasn't
>> strictly RAID10, to get a 2-way mirror out of three disks.  Let's call
>> the drives A, B, C.  All were the same size.  I split each drive into
>> two partitions, then mirrored the partitions like so:
>>
>> [ A1 | A2 | C1 ]
>> [ C2 | B1 | B2 ]
>>
>> then striped the three mirrors.  I had full redundancy in each mirror,
>> and no single drive failure could cause any mirror to fail.
>>
>> I don't know what the hell you'd properly call it.  :)
> _______________________________________________
> GEEKS:  http://www.sunhelp.org/mailman/listinfo/geeks
> 


-- 
  Phil Stracchino
  Babylon Communications
  phils at caerllewys.net
  phil at co.ordinate.org
  Landline: 603.293.8485