Convert raid5 to raid1?

Convert raid5 to raid1?

[John McMonagle]

Was planning to adding a hot spare to my 3 disk raid5 array and was 
thinking if I go to 4 drives I would be a  better off  as 2 raid1 arrays 
considering the current state of raid5.

If you think that is wrong please speak up now :)

Thinking I would  make a raid1  array for /.
The rest of the first pair  and the second pair of disks raid1 and tie 
together with lvm2.

Can easily fit what I have so far on 1 drive so figure I could build the 
first 2 raids on it  them copy from the current degraded array.

Just wonder what happens to the md sequence when I remove the original 
raid arrays?
When I'm done will I have md0,md1 and md2 or  md2,md3 and md4?

John

Re: Convert raid5 to raid1?

[Frank Wittig]

John McMonagle wrote:

> Just wonder what happens to the md sequence when I remove the original 
> raid arrays?
> When I'm done will I have md0,md1 and md2 or  md2,md3 and md4?

they will have the name you entered when you created the array.
after removing one array from the system all arrays will still have 
their original device-number after reboot.

greetings,
frank

Re: Convert raid5 to raid1?

[Brad Campbell]

John McMonagle wrote:
> Was planning to adding a hot spare to my 3 disk raid5 array and was 
> thinking if I go to 4 drives I would be a  better off  as 2 raid1 arrays 
> considering the current state of raid5.

I just wonder about the comment "considering the current state of raid5". What might be wrong with 
raid5 currently? I certainly know a number of people (me included) who run several "large" raid-5 
arrays and don't have any problems.

Brad
-- 
"Human beings, who are almost unique in having the ability
to learn from the experience of others, are also remarkable
for their apparent disinclination to do so." -- Douglas Adams

RE: Convert raid5 to raid1?

[Guy]

The only problem I have is related to bad blocks.  This problem is common to
all RAID types.  RAID5 is more likely to have problems.

Guy

-----Original Message-----
From: linux-raid-owner@vger.kernel.org
[mailto:linux-raid-owner@vger.kernel.org] On Behalf Of Brad Campbell
Sent: Thursday, March 10, 2005 6:04 PM
To: John McMonagle
Cc: linux-raid@vger.kernel.org
Subject: Re: Convert raid5 to raid1?

John McMonagle wrote:
> Was planning to adding a hot spare to my 3 disk raid5 array and was 
> thinking if I go to 4 drives I would be a  better off  as 2 raid1 arrays 
> considering the current state of raid5.

I just wonder about the comment "considering the current state of raid5".
What might be wrong with 
raid5 currently? I certainly know a number of people (me included) who run
several "large" raid-5 
arrays and don't have any problems.

Brad
-- 
"Human beings, who are almost unique in having the ability
to learn from the experience of others, are also remarkable
for their apparent disinclination to do so." -- Douglas Adams
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Convert raid5 to raid1?

[John McMonagle]

Brad

Not saying its broke.
Part of my reasoning to go to raid5 was that I could expand.
While it can be done I don't really see it as practical.
Also it's looking like I probably will not need to expand.

raid5 with 3 drives and 1 spare
or 2 - 2 drive raid1 drives have the same space.
Which is less likely to have a failure cause data loss?

I'm guessing raid1.
If I'm wrong I'd like to know now.

Also concerned about the resync times. It was going to take a couple 
days to resync under a rather light load if it weren't for the fact that 
it couldn't because of a bad drive and a kernel panic caused by the read 
error.
Still not certain about the cause of problem my current guess is the 
sata controller.

I'm glad there is work being done on the resync issue.
Also think the ideas to attempt to fix read errors are great.
My only suggestion is that there should be provision to send 
notification when it happens.
With both that would really help.

John

Brad Campbell wrote:

> John McMonagle wrote:
>
>> Was planning to adding a hot spare to my 3 disk raid5 array and was 
>> thinking if I go to 4 drives I would be a  better off  as 2 raid1 
>> arrays considering the current state of raid5.
>
>
> I just wonder about the comment "considering the current state of 
> raid5". What might be wrong with raid5 currently? I certainly know a 
> number of people (me included) who run several "large" raid-5 arrays 
> and don't have any problems.
>
> Brad

RE: Convert raid5 to raid1?

[Guy]

You asked:
"raid5 with 3 drives and 1 spare
or 2 - 2 drive raid1 drives have the same space.
Which is less likely to have a failure cause data loss?"

Assume 4 drives.

	With RAID5 using 3 drives and 1 spare...
======================================================
If a disk is kicked out because of a bad block, a re-build starts, no
problem here.  Data redundancy is lost until the re-build finishes.

If during the re-build, a second bad block exists on any of the 2 remaining
disk, then another disk is kicked out.  You array is now down.  The data can
be recovered, but it is tricky.

	With RAID 1+0, 4 disks no spare.
======================================================
If a disk is kicked out because of a bad block, no problem.  Data redundancy
is lost until the fail disk is replaced and a re-build finishes.

Now only 1 of the remaining 3 disks has high risk.  If that disk gets a bad
block, game over, the array goes off-line, however, the data is not really
lost and can be recovered, but still tricky.

==========================================================================

	With RAID 1, 1+0 and 5 if the 2 bad blocks occur on the same stripe,
then the data on the bad blocks is really gone, but that is very un-likely.

	So, IMO, RAID5 has a higher risk of going off-line do to multi block
(multi disk) failures.  RAID1 has less risk.  Assuming the same number of
disks.  Also, as the number of disks increases, the risk of failure on RAID5
goes up, and the risk on RAID1 goes down.

	RAID6 can survive 2 failed disks, very unlikely to have 2 more bad
blocks during a re-sync.  However, RAID6 with 3 disks and 1 spare is
useless.  But RAID6 with 4 disks should be more reliable than RAID1+0.  But
I am not convinced RAID6 is stable yet.  Others disagree with me.  I hope
they are correct.

	Raid1+0 and RAID10 are similar, but different.
RAID1+0 is a RAID0 array made up of 2 or more RAID1 arrays.  Each RAID1
array could lose 1 disk and the RAID0 array should continue to function.

RAID10 supports an odd number of disks.  In the case of an odd number of
disks, only 1 disk can fail, or the array goes off-line.  In the case of an
even number of disks, I am not sure how many disks can fail with out the
array going off-line.  RAID10 is also new, but I have not heard anything bad
(or good) about it.

Anything I say is my opinion.  My opinions are the best!  :)

Guy

-----Original Message-----
From: linux-raid-owner@vger.kernel.org
[mailto:linux-raid-owner@vger.kernel.org] On Behalf Of John McMonagle
Sent: Thursday, March 10, 2005 7:02 PM
To: Brad Campbell
Cc: linux-raid@vger.kernel.org
Subject: Re: Convert raid5 to raid1?

Brad

Not saying its broke.
Part of my reasoning to go to raid5 was that I could expand.
While it can be done I don't really see it as practical.
Also it's looking like I probably will not need to expand.

raid5 with 3 drives and 1 spare
or 2 - 2 drive raid1 drives have the same space.
Which is less likely to have a failure cause data loss?

I'm guessing raid1.
If I'm wrong I'd like to know now.

Also concerned about the resync times. It was going to take a couple 
days to resync under a rather light load if it weren't for the fact that 
it couldn't because of a bad drive and a kernel panic caused by the read 
error.
Still not certain about the cause of problem my current guess is the 
sata controller.

I'm glad there is work being done on the resync issue.
Also think the ideas to attempt to fix read errors are great.
My only suggestion is that there should be provision to send 
notification when it happens.
With both that would really help.

John

Brad Campbell wrote:

> John McMonagle wrote:
>
>> Was planning to adding a hot spare to my 3 disk raid5 array and was 
>> thinking if I go to 4 drives I would be a  better off  as 2 raid1 
>> arrays considering the current state of raid5.
>
>
> I just wonder about the comment "considering the current state of 
> raid5". What might be wrong with raid5 currently? I certainly know a 
> number of people (me included) who run several "large" raid-5 arrays 
> and don't have any problems.
>
> Brad


-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Convert raid5 to raid1?

[David Greaves]

John McMonagle wrote:

> Not saying its broke.

good :)

> Part of my reasoning to go to raid5 was that I could expand.

OK that is really quite hard.
Unless you run lvm2 over the top - in which case it's a doddle (though 
constraints apply)
I'd do it anyway (run lvm2). It causes almost no performance issues and 
you can then grow your fs later.
You could build a raid5 then linearly extend onto a raid0

> While it can be done I don't really see it as practical.
> Also it's looking like I probably will not need to expand.

Oh.

> raid5 with 3 drives and 1 spare
> or 2 - 2 drive raid1 drives have the same space.
> Which is less likely to have a failure cause data loss?
> I'm guessing raid1.
> If I'm wrong I'd like to know now.

A 4x raid5 gives 3X space and survives 1 disk failure (a spare can be 
added later)
A 2x raid1/raid0+2 gives 2X space and if 1 disk fails then there's a 1/3 
chance that a second failure will cause total failure.
A 3x raid5+1s gives 2X space and the ability for 2 simultaneous* disk 
failures
A 4x raid6 gives 2X space and the ability for 2 simultaneous** disk failures

* simultaneous here = apart from the few hours whilst a resync occurs - 
really I'm more interested in the lack of resilience during the few 
weeks whilst the RMA happens. Although I ended up with a spare 'cos I 
waited for a failure and bought a spare on next-day delivery whilst 
wrapping the failed disk for RMA

** really simultaneous

>
> Also concerned about the resync times. It was going to take a couple 
> days to resync under a rather light load if it weren't for the fact 
> that it couldn't because of a bad drive and a kernel panic caused by 
> the read error.

Hmm, took a few hours to resync 1Tb here. (maybe as much as overnight)

> Still not certain about the cause of problem my current guess is the 
> sata controller.

I run a 7 device SATA and it's81 stable (on 2.6.11.2 now).
I had a few minor problems a few months back but I think they've been 
sorted.

> I'm glad there is work being done on the resync issue.
> Also think the ideas to attempt to fix read errors are great.

Yes the "read error = kick disk" is still a significant failing :(

> My only suggestion is that there should be provision to send 
> notification when it happens

such as this one:

This is an automatically generated mail message from mdadm
running on cu.dgreaves.com

A Fail event had been detected on md device /dev/md0.

Faithfully yours, etc.


man madm
(or Debian sets it up automagically)

HTH

One other suggestion is to consider using lvm over 2 raid1 devices 
rather than md0/md1 - I think you'll find a lot more flexibility there.

David

Re: Convert raid5 to raid1?

[Paul Clements]

Brad Campbell wrote:
> John McMonagle wrote:

> I just wonder about the comment "considering the current state of 
> raid5". What might be wrong with raid5 currently? 

Perhaps he's referring to the possibility of undetectable data 
corruption that can occur with software raid5? Granted, there's a very 
small chance of it happening, but if your array becomes degraded (i.e., 
you lose a disk, which with the way md currently works, only takes one 
bad block) and then you have a server crash (or other unclean shutdown 
of your array), you likely will suffer data corruption.

See this for more details:

http://www.cs.helsinki.fi/linux/linux-kernel/2002-19/0906.html

--
Paul