Linux RAID 5 – Redundant Array of Inexpensive Disks

Linux RAID 5 – Redundant Array of Inexpensive Disks.

-- Tom

 

Friends don't let friends use RAID5

Here's what I said in https://www.ivpn.net/privacy-guides/advanced-privacy-and-anonymity-part-4/:

Cites:

http://www.reddit.com/r/sysadmin/comments/ydi6i/dell_raid_5_is_no_longer_recommended_for_any/

http://www.standalone-sysadmin.com/blog/2012/08/i-come-not-to-praise-raid-5/

 

We don't have a problem with RAID5, we have approx 50 servers with a minimum of 4 physical drives and a max of 16 drives, approx 1 to 15tb arrays, these machines are getting towards 5 years old. We get about 2 drives a month failing, some machines should in theory of suffered rebuild failure according various people's claims about RAID, but our real world experience is that the oft touted RAID5 is now useless appears to be scare mongering. The only single RAID failure we have had was not at rebuild time !
Maybe drives often perform better than error rates suggest (e.g. hardware error correction mitigates things), calculations maybe were for consumer grade drives where we use enterprise grade drives.
Then there is the fact we use hot spares, which build a lot faster with less reads if the controller replaces pre-failure, which our controllers often sense and adding a new drive in this case to RAID5 is as fast RAID10 as parity does not need to be rechecked/calculated (which invokes reading all drives across each stripe).

Write caching makes a more significant difference to RAID5 than RAID10, brought latency down to a level that was no longer a significant bottleneck for our work loads.

Yes RAID5 is more error prone on rebuild than RAID10, but RAID5 and RAID6 even more so are more resistant to errors than RAID10 during normal operation, parity offers error protection that RAID10 cannot offer.

From what I understand, RAID6 will always be more resilient to errors/failures than running same number of drives in RAID5, hot spare or not, due to the double parity, but at a greater performance penalty.

You need to decide on an individual basis what your priorities are, such as rebuild time, on line data integrity, performance, storage size, if you are running enterprise grade drives, what performance and reliability features your controller offers

 

@NGRhodes

Thanks for your professional perspective.

Your RAID5 setup using controllers that replace pre-failure with hot spares is very cool. You get the speed of and parity checking of RAID5, with reliability more like RAID10.

What RAID controllers are you using?

I wonder if something similar is possible in Linux software RAID.

 

We use mostly HP P400 controllers (on HP servers), they call it "Predictive failure".

 

What filesystem do you guys use?

Did anybody tested RAID-Z and ZFS filesystem on Ubuntu?

 

OS default, for best support.

 

I use ext4.

I've been meaning to look into that.

I find this <https://help.ubuntu.com/community/encryptedZfs>.