Best practice configuration?

[mir]

A raidz2 on ssd would eat ssd very fast due to write penalty. If i remember properly, for every write, a raid6 has to write it 3 times more

Not 3 times more but 3 times (data+parity*3). Exactly the same with your 3-disk raid1 (data*3). And we are talking enterprise DC SSD's here so they will not wear out before your normal recycle occurs.

What about recovery time? how long does it take to resilver a full ssd raid6 like the one wrote above?

Given sufficient CPU power my guess is that resilvering an all 300gb SSD raidz2 will be the same as for resilvering a 600gb sas mirror.

 

[Alessandro 123]

Why parity*3? Raid6 has 2 parity, so total writes should be 3: data+parity+parity

Why are you suggesting a raidz2 in opposite to a 3 disks raid10?

With my raid10 i have the same number of writes, no need to calculate parity, improved read performance, better fault tollerance and i can expand it by adding a new raid1, so i can start with just 3 disks and add 3 more disks when needed.

 

[mir]

Why parity*3? Raid6 has 2 parity, so total writes should be 3: data+parity+parity

I should have written: ((data/3)+parity)*3. To be able to support that any two disks can be lost parity needs to be spread over all 6 disks.

 

[Alessandro 123]

Ok but is my thinking still correct?
Why are you suggesting a raidz2 and not a raid1? The latter should be better in everything (if done like i wrote)

 

[mir]

Ok but is my thinking still correct?
Why are you suggesting a raidz2 and not a raid1? The latter should be better in everything (if done like i wrote)

I think you are wasting to much disk space for very little gain. If you loose 3 disks in the same vdev you will loose the entire pool anyway so not much more security. Also remember raid is not a substitute for backup!

 

[Alessandro 123]

That's not exactly true.
In a raidz2 you can loose any 2 disks on 6
In a raid10 with 3 disks on each mirror, you can loose 2 disks on 3, the probality is much lower.

 

[Alessandro 123]

Probably i'll go with a RAIDZ2 with 4 disks. Using 6 disks is too expesinve and making a RAID10 with 4 disks is much more insecure than a RAIDZ2 with the same usable space.

Any drawbacks? I've read somewhere that is suggested to use a multiple of 2 plus 2 disks for parity as optimal configuration, thus a RAIDZ2 with 4 disks would be ok (2^1+2). The next steps would be 2^2+2 (6 disks) or 2^3+2 (8 disks)

 

[mir]

Probably i'll go with a RAIDZ2 with 4 disks. Using 6 disks is too expesinve and making a RAID10 with 4 disks is much more insecure than a RAIDZ2 with the same usable space.

Any drawbacks? I've read somewhere that is suggested to use a multiple of 2 plus 2 disks for parity as optimal configuration, thus a RAIDZ2 with 4 disks would be ok (2^1+2). The next steps would be 2^2+2 (6 disks) or 2^3+2 (8 disks)

Only performance wise. RAID10 will give you higher IOPS and the requirements to the CPU are much lower since there is no parity calculations. Except for very few situations (Storage pools for DBMS and LHC) RAID6 is to be preferred over RAID10