Raid 5+1 or Raid 6?

[hokatichenci]

I'm stuck in sort of a pickle here. I've got 4 solid drives in raid 5. I just got one more in and I'm ordering the other 3 shortly for a total of 8 on my Areca 1220. Except that you can't convert from raid5->raid6 (downgrade only it seems). This leaves me with either expanding 3 more drives worth and using a hot spare (5+1) or loading 4 drives as raid6 (640gb) and copying all the data over, then deleting the old array, and expanding this array to RAID 6, which is quite a bit riskier I think.

I'm not really worried about the performance loss using raid 6, I'm mainly worried about losing all my data. Any input or suggestions?

 

[Vertigo Acid]

If you're really worried about your data, then what is your backup solution? The point of RAID 5 is uptime, about being able to throw in a drive if anothe goes south, not using it as an alternative to a backup

 

[hokatichenci]

All my critical data is backed up through CVS or on another drive thats not hooked up and is known to be good. I did however find out that you can actually upgrade an array to RAID6, however you have to have a spare drive first. So you have to expand+modify the array at once, or in my case not know you can do that and expand first, then modify. Its about 11% done after 30 minutes so in another 5 hours or so I should have a better idea of whether it works or not.

 

[[h]ardc[h]ris]

With raid6, you can lose up to 3 drives, before you lose data.

With Raid 5 + 1, if you lose a drive, you have to wait for the rebuild, and if you lose a drive during that, you've lost data.

 

[hokatichenci]

With raid6, you can lose up to 3 drives, before you lose data.

Fixed for accuracy against mix-ups: Raid6 works after two concurrent disk failures. Raid5 is one disk failure. I've already moved to RAID6 and posted some benchmarks somewhere else (~214mb/s on XFS).

 

[unhappy_mage]

Raid 6 allows for arbitrarily complex data+parity systems. As HPA puts it:

Reed-Solomon coding can be exploited further to allow for any combination of n
data disks plus m redundancy disks allowing for any m failures to be recovered.
However, with increasing amount of redundancy, the higher the overhead both
in CPU time and I/O. The Linux RAID-6 work has been focused on handling
the case of m = 2 efficiently in order for it to be practically useful.

I believe that the Areca card is also specifically locked to m=2. However, it's a general term. Just like raid 5 means any number of data disks and one parity disk, raid 6 means any number of data disks and any number of parity disks.

Now that I've made my point, yeah, it's two

 

[Nostradamus]

Fixed for accuracy against mix-ups: Raid6 works after two concurrent disk failures. Raid5 is one disk failure. I've already moved to RAID6 and posted some benchmarks somewhere else (~214mb/s on XFS).

Do you have a link to the benchmarks? I'd like to see that.

 

[Rocco123]

IMO RAID 6 is overkill. I would rather have one big RAID 5 array and get an extra disk. In 6 years, I have only seen one instance of a two disk failure. Only under a rare circumstance would I ever integrate RAID 6 into a server.

 

[Toytown]

IMO RAID 6 is overkill. I would rather have one big RAID 5 array and get an extra disk. In 6 years, I have only seen one instance of a two disk failure. Only under a rare circumstance would I ever integrate RAID 6 into a server.

In the enterprise environment you seem to see it more than you would like. Definetely more than once (2 disk failure) in 6 years. Over the past few years, ive seen countless power supply issues taking x amount of drives out and the RAID array having to be restored completely. This is with all different hardware vendors including mostly IBM / SUN.

Its definetely a case of keeping uptime figures high, but no replacement for a proper backup.