Does MD solve write hole issue for RAID5/RAID6?

Does MD solve write hole issue for RAID5/RAID6?

[rj]

Hi,
I wanted to know if md solves write hole issue for RAID5/RAID6?
Also, does md support RAID 50 level ?
Thanks!

Re: Does MD solve write hole issue for RAID5/RAID6?

[Asdo]

rj wrote:
> Hi,
> I wanted to know if md solves write hole issue for RAID5/RAID6?
>   
No but the filesystem can do that, ext3/4 and XFS in particular, if well 
aligned on the RAID, if you are creating new files and not updating old 
files (i.e. doesn't work for databases).
For full write hole avoidance you need to wait for Btrfs (and I don't 
know the state of raid-5/6 on btrfs), or use ZFS on Solaris or Freebsd.
> Also, does md support RAID 50 level ?
>   
Not directly, but you can create that manually by overlaying a raid 0 
over 1+ raid5's.

Re: Does MD solve write hole issue for RAID5/RAID6?

[Goswin von Brederlow]

Asdo <asdo@shiftmail.org> writes:

> rj wrote:
>> Hi,
>> I wanted to know if md solves write hole issue for RAID5/RAID6?

It is unsolvable without a crash persistant journal. Hardware raids have
battery backed cache for that but a fast disk would also work. But no
support for this in linux software raid.

> No but the filesystem can do that, ext3/4 and XFS in particular, if
> well aligned on the RAID, if you are creating new files and not
> updating old files (i.e. doesn't work for databases).
> For full write hole avoidance you need to wait for Btrfs (and I don't
> know the state of raid-5/6 on btrfs), or use ZFS on Solaris or Freebsd.

No raid5/6 in btrfs. Needs major restructuring of the on-disk data for
that.

ZFS on the other hand uses what they call raid-x. Which is a raid5 with
copy-on-write semantic. Any write to a virtual block will write to a new
physical block and update the parity to a new physical block too. Only
once that was written is the stripe atomically changed to the new
physical location. So no hole there.

There is also a zfs-fuse implementation for linux.

>> Also, does md support RAID 50 level ?
>>
> Not directly, but you can create that manually by overlaying a raid 0
> over 1+ raid5's.

And by layering one raid over others you can get any level you like,
even 14065 if you like (and have enought disks).

For raid50 I would suggest LVM over raid5 with striping though. Raid 0
is somewhat pointless when compared with all the extra flexibility LVM
gives you on top of striping.

MfG
        Goswin

Re: Does MD solve write hole issue for RAID5/RAID6?

[rj]

okay, thanks guys !

On Sat, Mar 6, 2010 at 1:02 AM, Goswin von Brederlow <goswin-v-b@web.de> wrote:
> Asdo <asdo@shiftmail.org> writes:
>
>> rj wrote:
>>> Hi,
>>> I wanted to know if md solves write hole issue for RAID5/RAID6?
>
> It is unsolvable without a crash persistant journal. Hardware raids have
> battery backed cache for that but a fast disk would also work. But no
> support for this in linux software raid.
>
>> No but the filesystem can do that, ext3/4 and XFS in particular, if
>> well aligned on the RAID, if you are creating new files and not
>> updating old files (i.e. doesn't work for databases).
>> For full write hole avoidance you need to wait for Btrfs (and I don't
>> know the state of raid-5/6 on btrfs), or use ZFS on Solaris or Freebsd.
>
> No raid5/6 in btrfs. Needs major restructuring of the on-disk data for
> that.
>
> ZFS on the other hand uses what they call raid-x. Which is a raid5 with
> copy-on-write semantic. Any write to a virtual block will write to a new
> physical block and update the parity to a new physical block too. Only
> once that was written is the stripe atomically changed to the new
> physical location. So no hole there.
>
> There is also a zfs-fuse implementation for linux.
>
>>> Also, does md support RAID 50 level ?
>>>
>> Not directly, but you can create that manually by overlaying a raid 0
>> over 1+ raid5's.
>
> And by layering one raid over others you can get any level you like,
> even 14065 if you like (and have enought disks).
>
> For raid50 I would suggest LVM over raid5 with striping though. Raid 0
> is somewhat pointless when compared with all the extra flexibility LVM
> gives you on top of striping.
>
> MfG
>        Goswin
>
--
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