bin/166589: atacontrol(8) incorrectly treats RAID10 and 0+1 the same

[Alexander Motin]

The following reply was made to PR bin/166589; it has been noted by GNATS.

From: Alexander Motin <mav at FreeBSD.org>
To: Allen Landsidel <landsidel.allen at gmail.com>
Cc: bug-followup at FreeBSD.org
Subject: Re: bin/166589: atacontrol(8) incorrectly treats RAID10 and 0+1 the
 same
Date: Tue, 15 Jan 2013 18:20:56 +0200

 On 15.01.2013 18:03, Allen Landsidel wrote:
 > I'm also extremely interested to hear how you intend to "handle it as
 > RAID10 at the OS level" since that is, in fact, impossible.
 
 Easily!
 
 > If it's a RAID0+1 in the controller, than it's a RAID0+1. Period.  The
 > OS can't do anything about it.  A single disk failure is still knocking
 > half the array offline (the entire failed RAID-0) and you are left with
 > a functioning RAID-0 with no redundancy at all.
 
 ataraid(8) in question (and its new alternative graid(8)) controls
 software RAIDs. It means that I can do anything I want in software as
 long as it fits into existing on-disk metadata format. If RAID BIOS
 wants to believe that two failed disks of four always mean failed array
 -- it is their decision I can't change. But after OS booted nothing will
 prevent me from accessing still available data replicas.
 
 > On 1/15/2013 10:55, Alexander Motin wrote:
 >> Their on-disk formats are identical. Even if RAID BIOS supports RAID0+1,
 >> there is no problem to handle it as RAID10 at the OS level. That gives
 >> better reliability without any downsides. I think there is much higher
 >> chance that inexperienced user will choose RAID0+1 by mistake, then
 >> experienced wish do to it on intentionally. Do you know any reason why
 >> RAID0+1 can't be handled as RAID10?
 >>
 >> On 15.01.2013 17:28, Allen Landsidel wrote:
 >>> Most devices typically only support one level or the other, but not
 >>> both.  I don't "Insist that it should exist", it *does* exist.  Both
 >>> levels do, and they are not the same thing.
 >>>
 >>> As for why it should be "available" to the user, I think that's a pretty
 >>> silly question.  If their hardware supports one or both levels, they
 >>> should be available to the user -- and called by their correct names.
 >>>
 >>> On 1/15/2013 03:12, Alexander Motin wrote:
 >>>> That is clear and I had guess you mean it, but why do you insist that
 >>>> such RAID0+1 variant should even exist if it has no benefits over
 >>>> RAID10, and why it should be explicitly available to user?
 >>>>
 >>>> On 15.01.2013 04:51, Allen Landsidel wrote:
 >>>>> They are not variants in terminology, they are different raid levels.
 >>>>> Raid0+1 is two RAID-0 arrays, mirrored into a RAID-1.  if one of the
 >>>>> disks fails, that entire RAID-0 is offline and must be rebuilt, and
 >>>>> all
 >>>>> redundancy is lost.  A RAID-10 is composed of N raid-1 disks combined
 >>>>> into a RAID-0.  If one disk fails, only that particular RAID-1 is
 >>>>> degraded, and the redundancy of the others is maintained.
 >>>>>
 >>>>> 0+1 cannot survive two failed disks no matter how many are in the
 >>>>> array.  10 can survive half the disks failing, if it's the right half.
 >>>>>
 >>>>> This is something people who've never used more than 4 disks fail to
 >>>>> grasp, but those of us with 6 (or many many more) know very well.
 >>>>>
 >>>>> On 1/14/2013 21:46, Alexander Motin wrote:
 >>>>>> There could be variants in terminology, but in fact for most of users
 >>>>>> they are the same. If you have opinion why they should be treated
 >>>>>> differently, please explain it.
 >>
 > 
 
 
 -- 
 Alexander Motin