I'm testing S2D now and I would like to calculate MTBF for the whole system. It seems very fragile to me:
Let's have 4 nodes (x) with 12 drives (y) in each. MTBF for one node is 1/12
of single drive value. For three-copy mirror we can tolerate 2 node failures (equals two failed disks in different nodes). That gives me 2/4
MTBF of single node value. In total it is 1/24
or 2/(x*y)
MTBF in comparison to single drive.
This value is quite low and gets even lower for larger deployments. Is my calc correct? I know that Microsoft recommends to reserve capacity of 2 disks in the pool aka hot-spares, but is it sufficient for production use?
Microsoft recommends % of usable capacity, "2 disks" is too bold statement for say 90 disk JBOD.