The STORNADO

[fohdeesha]

LTT has time on their hands for hundreds of hours, lol jk.

But based on the video and comment from Patrick, i believe there was a very slim chance of drive failure and being the drives are SSD it would resilver faster then HDD.

SSD is why I said hundreds of hours instead of months A 26 drive width vdev will resilver cripplingly slow - the big concern is with such a large time window and so many other drives, the probability of one or more failing in said time window and rendering the array useless/unrecoverable is not insignificant. certainly low, but not your usual zfs low. I know most ZFS diehards on.....certain forums lose their shit when they see raidz2 on anything more than 5 drives, let's hope they don't see this

 

[marcoi]

SSD is why I said hundreds of hours instead of months A 26 drive width vdev will resilver cripplingly slow - the big concern is with such a large time window and so many other drives, the probability of one or more failing in said time window and rendering the array useless/unrecoverable is not insignificant. certainly low, but not your usual zfs low. I know most ZFS diehards on.....certain forums lose their shit when they see raidz2 on anything more than 5 drives, let's hope they don't see this

it would be a fun test to see what would happen if someone had a play system similar to this build and they forced a failure. Plus it seems in this usecase they had access to a different branch of Freenas where they "optimized" for ssd, so who knows if that also includes resilver times.

 

[Evan]

Some commercial vendors are doing 18 drives this way, rebuilt about 4hr with 8tb ssd. (Granted it’s not sata drives)
Don’t forget ssd is suffering 1/10th or lower AFR compared to rust and often 2 orders of magnitude better URE (10x17 vs 10x15, and yes I know some rust does better than that)

 

[Patrick]

We have not updated our population stats since our last piece in 2016 https://www.servethehome.com/used-enterprise-ssds-dissecting-our-production-ssd-population/

Since then, we have added hundreds of SSDs. We have not seen a single production unit fail (we had a pre-production batch of SSDs where we saw failures but they never were on the market.)

Resliver Raid-Z1 is not the same as Z2. Also, SSD recovery speeds are much higher than hard drives. Granted, it would not be a great day if you had the array down and 12 people hammering it simultaneously. On the other hand, if you have say 3x 9 drive Z1's striped, you end up with not great performance during a rebuild as well.

We did not do Z3 because you have a higher chance of a PSU, RAID card, backplane or other component failure versus that many multiple disk failures.

You are right on a good idea to test. Since we have Rome next week and Flash Memory Summit, this is going to be some time out.