Parity Selection: Mirrors VS. Z1 in terms of risk

[sfcredfox]

There are many posts and a few experts that strongly discourage the use of Z1 because of the risk during re- silvering of a subsequent failure or errors that will render the pool INOP (If I understand correctly). They strongly encourage the use of Z2 for this reason.

For those who require performance, usually the appropriate pool configuration will be mirrors. So in this case, you still only have 1 parity disk in each set.

Discussion questions:
How is the risk different between a pool created with mirrors versus a pool created with Z1?

Is the risk of errors/subsequent failures reduced because the parity set is so much smaller (only two disks instead of potentially up to 11ish)?

I guess depending on the answer to the above question, are users accepting more risk with mirrors in order to gain performance?

I've seen some posts that insinuate that you can get decent performance,maybe even equal or close to it, out of Z1 sets instead of mirror. If you did small sets, would this be more acceptable against the risks discussed above instead of doing large sets?

Thanks.

 

[Scharbag]

I am sure there are many better explanations but the basic theory is this:

• Mirror requires both disk in the same vDev to fail for data to be lost.
  • let us say you have 5 vDevs of 2 drives each for 10 drives
  • up to 5 drive COULD fail without loss (as long as they were not in the same vDev)
• Z1 has only a SINGLE parity drive
  • so a 10 drive Z1 can ONLY survive a single failure
  • Probability of ANY other drive failing during resilvering is much higher than the probability of the second drive of the mirrored vDev failing

Mirrors are N/2 for efficiency. Z1 is N-1 for efficiency. It is all about the level of risk/reliability you are looking for.

My Z2 array made with 3TB drives takes over 20 hours to resilver. If it were a Z1 pool, any other failure in that 20 hours would pooch the entire pool.

Cheers,