[CentOS] OT: What's wrong with RAID5

[Christopher Chan]

Filipe Brandenburger wrote:
> Hi,
>
> On Thu, Sep 24, 2009 at 08:48, nate <centos at linuxpowered.net> wrote:
>   
>> RAID 6 has a pretty good performance hit vs RAID 5 due to the
>> extra parity disk, on some arrays the performance hit is even
>> greater as the array calculates parity twice, NetApp I think has
>> as good a RAID 6 implementation as there is, though they can't get around
>> writing the parity information to two different disks.
>>     
>
> Actually, NetApp tries to attenuate that impact.
>
> As the mirroring is coupled with the filesystem, NetApp uses an
> intelligent algorithm that tries to fill all the blocks in a whole
> stripe before writing to the disks, that way it only has to calculate
> the parity of data in RAM and can flush to all disks at the same time.
> That's why in most cases they do not have the "read before write"
> impact of RAID-5 and RAID-6. That's also why they have RAID-4 instead
> of RAID-5, they say that the parity disk will not be a hot spot as in
> most cases all the disks will be written at the same time. Well, at
> least that's what they say...
>
> I believe ZFS implements similar ideas. I guess that's one of the
> patents for which NetApp tried to sue Sun a couple of years ago when
> ZFS first came out.
>
>   
Not quite. ZFS does things the other way around. "RAID-Z is a 
data/parity scheme like RAID-5, but it uses dynamic stripe width. " From 
your description, NetApp massages data to fit the stripe. ZFS's raidz 
massages the stripe to fit the data.

http://blogs.sun.com/bonwick/entry/raid_z